Scaling AI for

Businesses in
Europe Advanced
manufacturing
industry
Grow with
Table of contents
1 Overview of trends in the advanced manufacturing industry and its
AI and automation potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 AI applications that can be implemented in the advanced . . . . . . . . .
manufacturing industry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1 Optimizing output by tuning production parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Optimizing production by forecasting demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Preventing phishing attacks by analyzing emails . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Optimizing manufacturing operations by using collaborative robots . . . . . . . . . . . . . 14
2.5 Improving product design and configuration by using AI-based
suggestions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Simplifying the procurement process by enhancing cross-functional
coordination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Optimizing quality control by visually detecting faulty parts . . . . . . . . . . . . . . . . . . . . 22
2.8 Improving return on marketing invest by optimizing marketing mix. . . . . . . . . . . . . . 24
2.9 Improving reliability at lower cost by using predictive maintenance. . . . . . . . . . . . . . 27
2.10 Preventing and detecting cyberattacks by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
analyzing threat data and network traffic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.11 Enhancing the efficiency of legal document generation and review
by using natural language processing and automation . . . . . . . . . . . . . . . . . . . . . . . . 32
2.12 Optimizing prices by identifying margin improvement and price
perception opportunities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.13 Optimizing production by forecasting demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.14 Enhancing the productivity of the pick-and-place process in
warehouses by using smart glasses with augmented reality . . . . . . . . . . . . . . . . . . . 40
2.15 Enhancing maintenance by using smart glasses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.16 Automating invoice processing by visually recognizing invoice data . . . . . . . . . . . . . 45
2.17 Enhancing the productivity of the pick-and-place process in
warehouses by using smart glasses with augmented reality . . . . . . . . . . . . . . . . . . . 47
2.18 Increasing workers’ safety by using machine vision. . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2
2.19 Automating procurement contract management by using semantic
text analytics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.20 Automating lead scoring and prioritizing customers by analyzing
the likelihood to buy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.21 Automating expense accounting by using visual recognition . . . . . . . . . . . . . . . . . . . 56
2.22 Improving project performance by predicting success factors . . . . . . . . . . . . . . . . . 59
2.23 Improving HR processes by using HR bots to automate
employee interactions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.24 Optimizing payments by analyzing invoices with natural
language recognition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.25 Automating CV screening by using visual recognition . . . . . . . . . . . . . . . . . . . . . . . . . 66
3 Appendix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1 Cloud computing offers SMEs the possibility to dynamically
and economically scale computation, storage, and network capacities . . . . . . . . . . 69
3.2 The General Data Protection Regulation (GDPR) protects personal
data and allows the free flow of data within the EU. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4 Sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

3
1 Overview of trends
in the advanced
manufacturing
industry and its AI and
automation potential
The advanced manufacturing industry accounts for approximately 6 percent of the total
European value added and has a comparably low SME share (26 percent) (Eurostat, January
2020; Eurostat, 2016).1 Advanced manufacturing is a diverse industry, with manufacturing
of transport vehicles as the largest subindustry (40 percent), followed by machinery and
equipment (32 percent) as well as the manufacturing of computer, electronics, and electrical
equipment (28 percent). Since this industry encompasses high-tech, research-intensive
manufacturing, it has thus not been affected by the structural shift from manufacturing
towards services to the same extent as basic manufacturing. Advanced manufacturing
developed favorably in the EU, with an average annual growth of real value added of 3.0 percent
between 2000 and 2017, twice the rate of the total economy of 1.5 percent.2 The advanced
manufacturing industry quickly recovered from the 2008 financial crisis, with value-added values
surpassing the 2007 level already in 2011. Between 2014 and 2017, growth accelerated to 7.9
percent, which was the highest value across industries in this period. Whether this trend will
hold in the future in terms of real growth is uncertain. It largely depends on the extent to which
the recent trend of technological progress can be sustained, which would be reflected in further
decreases of quality-adjusted prices. Nominal growth, which does not adjust for technological
progress embedded in products, is expected to reach 2.0 percent until 2025, on par with the
long-term historical value of 2.1 percent (IHS, February 2020). Globalization has significantly
influenced the industry: while the efficiency of supply chains has greatly improved, international
competition has become stronger. As investment is shifting towards intangible assets such as
R&D, the costs of innovation are rising compared to value added (MGI, February 2018). These

1 When not explicitly stated otherwise, data in this paragraph is taken from Eurostat’s national accounts by industry database.
2 Growth rates are always indicated as CAGR.

4
forces have led to an ongoing consolidation of activities, e.g., in the automotive and advanced
electronics subindustries.
The technical automation potential includes image/sound recognition, machine learning,
and other AI technologies and is based on currently demonstrated technologies. The
technical automation potential of the advanced manufacturing industry is the highest of all
20 industries into which the economy is divided, on par with basic manufacturing.3 This is
due to the high prevalence of predictable, repetitive physical activities. 4 However, in terms of
digitization, advanced manufacturing is better positioned to start using AI at scale than basic
manufacturing. The digitization level is in the upper quartile of the ranking (Rank 4 out of 20),
while basic manufacturing is in the lower half, both in terms of the general index and across
all three pillars of digitization.5 Advanced manufacturing tends to exhibit higher digitization
of assets and labor compared to basic manufacturing, e.g., in digital spending on hardware,
software, and IT services (both in share of expenditure and per worker). There is an even larger
difference in the usage of digital tools in transactions (online sales and purchases), interactions
with suppliers and customers, and business processes (enterprise resource planning and
customer relationship management) (MGI, June 2016).

In advanced industries, AI has the largest potential in production optimization (e.g., yield, energy,
or throughput) and predictive maintenance. Furthermore, AI can help improve quality (e.g.,
through visual in-line quality controls). Other AI applications can optimize procurement (e.g.,
through spend analytics), product design (e.g., through AI-based configuration suggestions), or
back-office operations (e.g., through payment automation).

3 The 20 industries are based on the NACE industry classification from Eurostat (European System of Accounts 2010).
4 For more information on the methodology behind the technical automation potential and the different types of tasks and jobs,
see MGI (January 2017) and MGI (December 2017).
5 The MGI Industry Digitization Index is based on (1) the share of digital in spending and assets in terms of hardware, software,
IT services, and telecommunications, (2) digital labor (digital spending and assets per worker and share of digital jobs), and (3)
digital usage, i.e., transactions (online sales and purchases), interactions with suppliers and customers, as well as processes
(enterprise resource planning and customer relationship management).

5
2 AI applications that
can be implemented
in the advanced
manufacturing industry
2.1 Optimizing output by tuning production
parameters
Description:
The goal of this AI application is to increase high quality output of a production process by
analyzing production data, finding factors that influence the quality and quantity of the output
and optimizing them. Given the large number of influencing factors, from environmental
parameters (e.g., temperature and humidity) to machine settings (e.g., pressure, voltage) and
input quality (e.g., sizes, batch variability), it is often impossible to deduct the reason for a
change in output quality. To tackle this complexity, AI can make influencing factors visible. The
AI solution tries to find correlations between (combinations of) input and output variables. E.g.,
it might find that production on hot summer days with a certain humidity level yields inferior
output when pressure levels are not adjusted. This AI application is especially interesting for
SMEs who do not have the production optimization resources (e.g., engineers dedicated to
this) that larger players can use. The AI solution can give hints for improvements without prior
knowledge our prejudices.

Economic impact:
By optimizing production, SMEs can expect a 3 to 7 percent increase in high quality output
(expert interviews). This depends on the starting point (e.g., is there already an optimization
scheme in place), the production complexity (e.g., for a complex production process which
is influenced by many parameters, the potential is higher) and the data that can be used for
analysis: The more data is available, there more likely is a higher potential.

6
Availability of off-the-shelf solutions:
There are off-the-shelf solutions available (e.g., IBM Maximo, ABB) which do require some
degree of customization from the company’s side. This depends on the machines used, the data
measured already and the available interfaces.

Data availability:
As this AI application tries to find correlations between input parameters and outputs, two types
of data are needed. Firstly, to optimize production, one needs to define the desired output. The
output quality has to be measured (e.g., from images, sounds or other sensor data) and then
assessed, (e.g., by measuring a percentage value of ideal quality). Secondly, data related to the
production process (the input parameters) is necessary. This can encompass three types of
data: Environmental data (e.g., temperature, humidity), machine data (e.g., pressure, voltage)
and data on the production inputs (e.g., size, quality). Of these, machine data is often available
as part of the normal operation of the machine. However, it may need to be digitized and sent
over a network and sometimes additional sensors need to be installed to capture required data
points (e.g., a machine may not measure temperature, or the measurement may not be granular
enough). Both environmental data and inputs’ data are not essential for the optimization to
work, but are likely to enhance results. They have to be measured with additional sensors,
digitized and connected, if they cannot be sourced from external vendors.

Technological infrastructure:
Sensors for output data (e.g., cameras) as well as production data other than machine data
(e.g., room temperature) are needed. Measured data needs to be digitized and a network
structure or cloud connection is required to upload them into the AI engine. The AI algorithm
can run both on proprietary computer hardware or cloud infrastructure.

Capital requirements:
Proof of concepts start at EUR 1,000 (IBM, n.d.). However, full-scale production optimization
systems are much more costly, with the exact expense strongly depending on the types of
sensors, equipment, and software needed. The more data has already been collected and can
be reused, the less investment is required. For example, if there is an existing visual quality
control and machine data can be tapped into, the data “only” needs to be connected and
analyzed.

7
Talent requirements:
While implementation may be feasible with external experts or employees of the equipment
provider, expertise is required in interpreting the results for daily operations: (e.g., which
recommendations should be acted upon?). This requires a basic understanding of the
optimization algorithm (e.g., how are recommendations generated). Thus, some training of
existing staff is necessary.

Organizational and cultural challenges:

To ensure maximum impact of this AI solution, it is necessary to onboard employees that
manage and operate the production process. They need to understand how the AI application
can help them and how they should react to the resulting recommendations (e.g., not every
recommendation may be feasible as other constraints (like engine speed) may apply). This
requires training and overcoming of potential trust issues (e.g., fear of AI).

Job impact:
The introduction of this application could have a slightly positive impact on jobs. Increases in
quality and quantity might entail a small increase in marketing, sales, and logistics work.

Environmental sustainability:
An indirect positive environmental effect is expected. Optimizing the production output is likely
to reduce waste, as the share of output with insufficient quality is reduced.

Examples:

• A polymer producer achieved yield increase by 3 percentage points while reducing specific
energy consumption by 12 percent. (expert interviews)

2.2 Optimizing production by forecasting demand

Description:
Predicting demand enables producers to make the right product in the right quantity at the right
time. This can increase overall sales as demand can be met during peak times that would not
be met without proper prediction. Furthermore, inventory levels can be reduced when inventory
is adapted to production needs. While traditional demand forecasting models are mainly
based on historical sales, AI can help to uncover non-linear patterns (e.g., dependencies on
weather). The AI solution tries to find correlations between combinations of input data (e.g.,

8
weather, demographics) and resulting sales. It might, e.g., find that good weather on certain
days combined with holidays in a certain region leads to increased sales. Demand forecasting
is especially important for SMEs to remain competitive as their production often does not profit
from the same scale effect that larger companies leverage.

Economic impact:
AI-based approaches to forecasting can reduce forecasting errors by 30 to 50 percent in
some settings. Furthermore, lost sales due to product unavailability can be lowered by up to
65 percent. Costs related to transport and ware-housing, and supply chain administration are
expected to decrease by 5 to 10 percent and 25 to 40 percent, respectively. Due to AI, overall
inventory reductions of 20 to 50 percent are feasible. (McKinsey, April 2017)

Availability of off-the-shelf solutions:

There are off-the-shelf solutions ((e.g. Relex, Blue Yonder, Lokad) available which do not require
further customization from the company’s side.These have to be integrated into business data
systems by technical consulting staff.

Capital requirements:
There are two types of costs associated with this AI application: The costs of the software
itself (roughly EUR 20,000 to 40,000) and the cost of the implementation, ranging roughly from
EUR 20,000 to 50,000 (Demand-Planning.com, n.d., Silvon.com n.d.). The implementation effort
depends on the sophistication and complexity of existing data management systems. For
example, if there is one consolidated CRM system with an accessible data interface, integration
will be relatively straightforward. If more internal systems of the company (e.g., ERP system) are
connected set-up will become more complicated as additional interfaces will be needed.

Data availability:
To forecast demand, two types of data are needed: sales data (e.g., orders fulfilled, orders
missed) and potentially influencing parameters (e.g., location, weather, demographics). Sales
data are typically available within customer relationship management systems (CRM). Other
data (e.g., weather) is typically either publicly available or can be bought from specialized data
providers.

Technological infrastructure:
There are no high computational requirements associated with the application. The software
runs as a software as a service (SaaS) application on any desktop computer and requires no
further technological infrastructure (apart from an internet connection).

9
Talent requirements:
Implementation of off-the-shelf solutions is feasible without any specific in-house analytics. The
application can run without the need for AI experts. However, data scientists and engineers can
help to reduce implementation efforts and continuously adapt the solution, to new scenarios.

Organizational and cultural challenges:

No major changes are required when existing demand forecasting schemes are enhanced
with this solution. Even if the organization starts the demand planning from scratch, no major
changes are required. However, depending on the variability of forecasts (e.g., a large change in
quantity), a closer integration with procurement and production planning may be necessary to
reap the full benefits of this AI application.

Job impact:
The introduction of this application could have a slightly positive impact on jobs. If proper
demand planning results in increased sales, additional jobs might be created across business
functions (e.g., in production, marketing, sales, logistics)

Environmental sustainability:
An indirect positive environmental effect is expected. If left-over production and inventory can
be reduced, this saves resources, leading to an indirectly positive environmental impact.

Examples:

• A Brazilian personal care producer employed demand forecasting and reported a 27 percent
reduction in shortage (sales, that cannot be fulfilled) and 22 percent reduction in inventory.
(expert interviews)
• A consumer packaged goods company improved forecast accuracy to 95 percent three
months into the future starting from 83 percent 10 days into the future. (expert interviews)
• OTTO, a German retail player, introduced an AI system to predict sales of the next 30 days
with 90 percent accuracy. (The Economist, April 2017)

2.3 Preventing phishing attacks by analyzing emails

Description:
Digital attackers try to steal information (e.g., account logins) by sending phishing emails that
look harmless or seem to come from a legitimate source. Phishing attacks pose significant
risks to businesses, for example, by leading to denial of services (DOS) attacks that block an

10
IT system with too many requests, introducing ransomware that encrypts valuable data until
a ransom is paid, or exposing sensitive data (e.g., customer data), which may lead to brand
damage or the payment of fines. The increasing complexity of the attacks makes it difficult for
existing rule-based systems to detect them. For example, “polymorphic” phishing campaigns
avoid detection by continuously changing the email that is being sent (e.g., sender name,
subject, email body, attachment file size) (Ironscales a, n.d.). Another attacking technique,
“spear phishing”, targets a specific individual such as a CFO by leveraging AI-enhanced social
engineering, e.g., collecting information from social profiles and other online sources. AI can
help companies protect themselves against such advanced threats by continuously learning
to identify suspicious patterns from combinations of dozens of email features, e.g., content
and context of email, webpage URL, or other obfuscation techniques (Vade Secure a, n.d.). As
soon as a threat is discovered, the AI solution pulls out of all affected email inboxes within an
organization. Additionally, the solution continuously trains employees by sending mock phishing
emails (Ironscales b, n.d.). While phishing poses risks to all businesses, SMEs are especially
vulnerable since they are often less aware or capable of defending themselves against these
attacks (Belami, June 2019). Therefore, AI-powered off-the-shelf solutions could be a solution
for SMEs.

Economic impact:
The economic impact lies in avoiding risk. For example, two-thirds of UK-based SMEs reported
cyberattacks in 2018, which caused damages (i.e., damaged assets, payment of fines, business
downtime) of EUR 80,000 on average, the most common cause being phishing attacks.
(Ashford, March 2019). For SMEs that aren’t investing enough in cybersecurity due to difficulties
in attracting IT talent (Smallbusiness.co.uk., March 2019), this AI application could help reduce
their risk exposure without having to recruit specialists.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions (e.g., Ironscales, Vade Secure, Barracuda).available which
do not require further customization from the company’s side. Solutions are available both as
on-premise and cloud-based email apps and can be integrated in existing email software (e.g.,
Microsoft Office).

Capital requirements:
Prices for this solution usually depend on the number of users and the type of service level (e.g.,
detection only, training of employees) and therefore vary between EUR 3 and EUR 5 per user per
month (Expertinights.com, n.d.).

Data availability:
This application requires only email data. Off-the-shelf solutions come readily trained to detect

11
patterns from dozens of unique features of the email to identify phishing, e.g., content and
context of emails, webpages, URLs including any redirects, URL shorteners, or other obfuscation
techniques (Vade Secure a, n.d.).

Technological infrastructure:
This application is available both as on-premise and cloud-based email apps. It requires access
to companies’ email systems (e.g., Microsoft Office) via an interface (API) and runs on any
computer or in the cloud.

Talent requirements:
Off-the-shelf solutions allow full outsourcing of phishing security (detection and reaction)
to the service provider. The integration into existing email systems (e.g., Microsoft Office)
via a preconfigured API takes less than an hour and can be performed by local IT staff
(Expertinsights.com. n.d.).

Organizational and cultural challenges:

No organizational and cultural challenges are expected from the application as its impact is
filtering out malicious emails. Some solutions offer an anti-phishing training by generating mock
phishing emails to train employees on reporting them (Ironscales b, n.d.).

Job impact:
The potential impact on jobs is unclear. On the one hand, AI helps reduce cybersecurity-related
work, thus leading to lower labor costs. On the other hand, many SMEs do not have dedicated
cybersecurity employees, so labor is not impacted.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Gandi, a French hosting and email provider, deployed an AI-based phishing defense solution
to increase the rate of detecting risky emails by a factor of 20, compared to the previous
solution.

• A US-based PR agency deploys an AI-based phishing defense solution, which automatically

detects 99 percent of the attacks (Ironscales c, n.d.).

• An international construction company used AI-based phishing defense when traditional

12
 systems failed to protect them against a polymorphic phishing attack by continuously
changing its (e.g., file size, subject, address). Having verified the message as malicious, the
application connected to all mailboxes and automatically removed the malicious message
enterprise-wide. This prevented all employees from opening further messages and stopped
the attack in its tracks. (Ironscales a, n.d.).

Intergamma, a Dutch DIY chain, deployed AI-based phishing defense and successfully detected
sophisticated attacks that traditional solutions failed to identify (Barracuda, n.d.).

2.4 Optimizing manufacturing operations by using

collaborative robots
Description:
Collaborative robots (“cobots”) work alongside, interact, and assist human workers in
manufacturing setups. Typically, they consist of a robotic arm which can be equipped with
various tools (e.g., for gripping, screwdriving, or gluing). Cobots are easy to program, relatively
inexpensive, and designed to work alongside humans ensuring maximum safety (Robotics and
Automation, 2018). AI-powered cobots can recognize the presence and orientation of parts
and perform dynamic pick-and-place tasks, read results from testing equipment and make
decisions accordingly, adjust the force control required to pick parts from a stack, detect and
evade collisions, or respond to errors with retry strategies (Universal Robots, 2019). This makes
them easy to deploy as they can learn by demonstration (i.e., the worker shows the robot the
task by guiding its arm) and optimize their behavior subsequently. They can take over a wide
range of tasks (e.g., machine tending, pick-and-lace, assembly, quality inspection), many of
them being repetitive and injury prone. This helps free up employees to concentrate on other
tasks that require creativity, solution-oriented thinking, and initiative. Cobots are especially
interesting for SMEs with high product diversity and different low-volume production processes
with varying end products (e.g., in term of batch size, product type). Employing industrial robots
in this context is difficult as they are designed to continuously perform repetitive tasks with
high precision. Changing their task is time consuming and requires significant reprogramming.
In addition, the purchase of robots is expensive, and programming and installation takes a lot
of time and specialized skills. Robots also tend to take up a lot of space and in order to not
endanger people on the shop floor, they have to be covered by safety screens. Cobots, on the
other hand, are designed to be used in SMEs: they are relatively cheap and easy to set up, can
be reallocated to new tasks within a day, and work alongside human workers.

13
Economic impact:
Implementing an AI-powered cobot in an SME manufacturing setup can significantly improve
productivity of the plant as individual processes can be improved by up to 200 percent (Fifth
Third Bank, n.d.). This frees up workers for other tasks. The average payback period can be as
low as less than a few months (Universal Robots, n.d.).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions (e.g., Universal Robots UR series, ABB YuMi, Automata Eva,
Rethink Robotics Baxter and Sawyer, Franka Emika, Kuka LBR iiwa, Fanuc) available which do
require some degree of customization from the company’s side. By using the user friendly and
intuitive displays of the cobots, the internal engineering workforce should be able to (re)program
them as required. Cobots are also available with additional features (e.g., added 3D visual
capability, motion control, path planning toolkit) as per the requirement of the specific shop floor
application.

Capital requirements:
The price of a cobot depends on payload capacity, reach, and additional features (e.g., added
3D visual capability, motion control, path planning toolkit) required by the specific shop-floor
application. The average cost for a cobot is approximately EUR 32,000 (with a payload of 5 kg
and a reach of 850 mm) and is all inclusive, which means no further accessories are required
to make it work. However, prices can be significantly higher at EUR 60,000 to 90,000 for more
sophisticated models (The Robot Report, 2017).

Data availability:
Cobots are typically trained by demonstration (i.e., the worker shows the robot how to
accomplish a task and then corrects its actions). No additional data (e.g., on the production
process) is needed for this. However, individual subtasks (e.g., recognition of different parts)
may require further data (e.g., labeled images). Once operational for a specific application or
use case, cobots do not require any external data input to function.

Technological infrastructure:
Besides the actual cobot (hardware), little to no additional technological infrastructure is
required. Most cobots can be deployed in less than a day and allow for quick changeovers with
minimal setup time and effort. With no fencing required, cobots can be moved around the shop
floor on wheels to complete various tasks.

Talent availability:
The talent requirement is relatively low. The engineering workforce can be trained to program

14
the cobot to accomplish the required tasks. The displays provided are user friendly and intuitive,
which makes is unnecessary to hire a special robotics team for the purpose. Deployment can
often be accomplished in less than a day (Robotics and Automation, July 2018).

Organizational and cultural challenges:

Cobots are designed to work with operators instead of replacing them, hence positive
perception by employees is to be expected. However, at the time of introduction, operators may
feel their jobs are threatened. Nevertheless, once employees are relieved from dirty, unsafe,
boring, monotonous, or repetitive tasks by the cobot, they are likely to perceive it more positively
and might even enjoy their work more. In this sense, cobots contribute to an attractive working
environment (Wired Workers, n.d.).

Job impact:
The introduction of this application could have a slightly negative impact on jobs, as less
operators will be required on the shop floor once cobots are installed.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Task Force Tips, a US-based manufacturer of firefighting equipment, employed cobots to

tend CNC machines, freeing up workers for more complex tasks. This reduced staffing
requirements from seven to three operators and resulted in a payback period of about one
month for the robot (Universal Robots, n.d.).

• ZEISS Metrology Centre, part of Carl Zeiss India (Bangalore) Pvt. Ltd, installed a Universal
Robots UR5 cobot that tends three metrology machines in the same cycle, replacing
the earlier manual loading systems that were problematic and inflexible when changing
production lines. Their use was also limited to the presence of staff with only one operator
being able to operate a single machine at a time, thereby limiting productivity. The UR5 gave
Carl Zeiss India the flexibility to integrate three machines through a single cobot, resulting
in an increase in machine utilization from 50 to over 90 percent. The cobot also allowed for
adaptability through seamless, problem-free switches in the production line that can easily
be accomplished in one or two days, making 24-7 operations a reality (Cobots.ie, 2019).

• InPrint, a Danish printing company, employs a cobot to pick and place print plates, reducing
nonproductive wait time while improving the work environment and freeing up two labor

15
 hours in capacity per day (Universal Robots, n.d.).

2.5 Improving product design and configuration by

using AI-based suggestions
Description:
Traditionally, engineers have been responsible for interpreting the design requirements, coming
up with ideas to fulfill the requirements, and sketching, drawing, and tweaking the design
manually to come up with a final version, which may or may not be optimal. This process can be
automated through AI algorithms that can generate product designs that optimize predefined
target parameters (e.g., reduce weight of key components). Generative Adversarial Networks
(GANs) are a deep learning technique which combines two neural networks, one which creates
new designs optimizing a set of key parameters and another which validates the generated
design on meeting target requirements. The latter neural network learns valid designs patterns
and reduces all possible versions to the relevant ones. Thus, the power of AI lies in the speed at
which it can generate new designs, test them, and suggest the most promising product designs
optimized with regard to certain parameters. For SMEs, this is particularly relevant because it
allows them to achieve optimized product design faster with a lean product development team.

Economic impact:
Adopting an AI-based methodology can result in optimized product designs achieved at
expediated speed. On average, 40 to 50 percent weight reduction of key components and
20 percent strength increase (Modern Machine Shop, 2019) can be accomplished with 20 to
25 percent average reduction in project lead time (expert interviews). This can be achieved with
a lean product development team and thereby lower labor cost.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available (e.g., Altair Inspire, Autodesk Fusion360, ANSYS
Discovery, Desktop Metal Live Parts, MSC Apex, nTopology nTop, PTC Creo, Paramatters
CogniCAD) which do require some degree of customization from the company’s side. While
off-the-shelf solutions can be deployed for generative design, companies need to work with
the solution provider to customize and execute the solution as per their product development
requirements (e.g., specific target parameters).

Data availability:
Three types of data are needed for the application, target parameters, historical product

16
designs, and technical requirement of components. The target parameters (e.g., target weight,
strength, size) can be set internally or, in some cases, sourced from user feedback (e.g.,
historical product feedback or specialized surveys can be conducted to understand user
requirements) with direct implications for user experience (e.g., seat design). Historical product
designs can be used as training data along with technical requirements of components (such as
grade of material used, manufacturing options).

Technological infrastructure:
An interface may be needed between the off-the-shelf solution and company’s existing CAD
or alternative design software for accessing historical designs, as required. Otherwise, most
solutions can be run on any desktop PC or in the cloud. Once the solution is procured, it can be
set up for use by the off-the-shelf provider within a week’s time.

Capital requirements:
The implementation costs for generative design vary significantly and are driven by the costs
for the off-the-shelf solution, integration with internal design software, and quantum of activity
performed on the solution. Subscriptions cost anywhere from EUR 450 to EUR 2,300 per year
with the scope of the solution ranging from only product design to complete workflow, including
product design, documentation, and manufacturing design (engineering.com, 2018).

Talent availability:
Despite availability of solutions from external providers, translation of design requirements
into AI algorithm inputs needs to be done in a structured way and with expert knowledge
(e.g., specific number of fixtures required by a component translated into design constraints).
Additionally, the engineers/designers need to be qualified to use the output of the AI algorithm
in the right way and approve the most suitable designs for testing. While AI will help speed up
the process, the final design needs to be confirmed by well-informed human decision makers.
Therefore, qualified talent (e.g., engineers) is required to implement this AI application. However,
as the typical users of AI-based product design have been (manual) designers before, they can
be trained and upskilled to cover the new tasks.

Organizational and cultural challenges:

Initially, there might be reluctance to accept these generative design solutions due to the
displaced labor. In addition, trust in the AI’s ability to compete with human designing capabilities
needs to be built. However, adoption is likely if the benefits of expediated and optimized design
output with lower efforts are realized by the team. It is to be noted that this acceptance is
absolutely critical to reap the benefits of AI integration in the product design development
process.

17
Job impact:
The introduction of this application could have a negative impact on jobs, as the number of
engineers/designers required for product development can be reduced with help from AI.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Airplane manufacturer Airbus used AI (generative design) to reimagine an interior partition

for its A320 aircrafts and came up with an intricate design that ultimately shaved off
45 percent of the weight off the part and also proved to be stronger than the original
partition design. This weight decrease is estimated to result in a massive reduction of jet
fuel consumed and a reduction of 465,000 metric tons of CO2 emissions per year (New
Equipment Digest, 2018).
• In 2018, at auto manufacturer General Motors’ (GM) Tech Center in Warren, Michigan, GM
and Autodesk engineers worked together using generative design to create 150 new design
ideas for a seat bracket – a small but important part used to ensure seat belts are fastened.
The design options are based on goals and parameters such as weight, strength, material
choice, fabrication method, and more. The selected new part turned out to be 40 percent
lighter and 20 percent stronger than the original part (Autodesk 2018).

18
2.6 Simplifying the procurement process by
enhancing cross-functional coordination
Description:
A large share of the work in procurement is spent on coordinating needs, requirements, and
orders between different departments. For example, the sales department might expect more
sales of a certain product, which requires more production, and therefore more supplies. Those
need to be stored, which needs to be cleared with logistics. While people and processes are
usually adapted for direct procurement (inputs for production), this coordination problem
is even more pronounced for indirect procurement such as IT services and travel. AI can
analyze data from various systems (e.g., enterprise resource planning (ERP) and learn to make
suggestions when one department has a new or changed demand. One way to implement this
would be with a convolutional neural network that learns which inputs are required for a given
demand. For example, it can deduct from previous experience that an increase in sales of a
product (e.g., a certain car seat) by x percent means that certain raw materials (e.g., leather)
need to be ordered y weeks in advance in quantity z. This application is relevant for SMEs who
often rely on personal relationships between employees rather than standardized processes.

Economic impact:
In a typical advanced manufacturing business, 15 to 25 percent of procurement activities are
focused on cross-functional topics (e.g., communicating needs and requirements). These
can be reduced by around 30 to 50 percent through the AI solution. Furthermore, capacities in
other functions like production and logistic sales are also freed up by roughly the same amount
(expert interviews).

Availability of off-the-shelf solutions:

There are currently no off-the-shelf solutions available for this application. A solution needs to
be custom-built for the respective company.

Data availability:
For this AI application to yield a maximum benefit, integrating all relevant data pools is
essential. These include data from procurement, warehousing, production, and sales. Often,
these data are available if an ERP system exists. Otherwise, they must be collected and digitized
for this application to work.

Technological infrastructure:
Given the large quantity of data and necessary computation power, this application needs to be

19
run in the cloud – this requires a fast internet connection. On premises, no further technological
infrastructure is required, as it can be accessed from any desktop computer.

Capital requirements:
As no off-the-shelf solutions are available, this application needs to be custom-built for the
business. The costs are driven by the capital expenses for a data scientist with experience
in multivariate modeling. The data analysis can be performed with open programming
environments (e.g., Python and R) and using open-source code (e.g., Keras). Typically, a team
of 2 to 3 data scientists needs between 6 and 20 weeks, depending on the existing data, the
complexity of existing systems, and processes. This would cost roughly between EUR 30,000
and EUR 130,000, assuming a data scientist’s salary of around EUR 55,000 (PayScale, n.d.) and
nonwage labor costs in the same amount.

Talent requirements:
To implement this solution, a data scientist with experience in multivariate modeling is required.
Once operational, the application can run on its own. However, new integrations are required
when the inputs change (e.g., IT system updates, or new product launches).

Organizational and cultural challenges:

Especially in SMEs, personal relationships between employees are often more important than
processes. Having an AI tool recommending procurement orders may be difficult for employees
to accept at first, but as soon as the simplifications become clear and reduce workload,
adoption is likely to be relatively straightforward.

Job impact:
The introduction of this application could have a negative impact on jobs, as procurement labor
time is significantly reduced.

Environmental sustainability:
No direct or indirect environmental effects are expected.

20
2.7 Optimizing quality control by visually detecting
faulty parts
Description:
Quality control is a key function in advanced manufacturing, as defects can lead to increased
costs (e.g., returns) and to human accidents and loss of business (e.g., faulty plane
components). Automated quality assurance systems are useful to ensure consistent quality
and enable a high-speed production. One example is a visual quality inspection system for
a production line of various products such as printed circuit boards or car bodies. Current
automated approaches for visual inspection take a picture of the product whose quality
should be checked and compare it to an ideal reference image of a product pixel by pixel. Such
systems typically require intense upfront investment and extensive testing and calibration.
This methodology is only feasible, however, if task- and environment-related preconditions
are met. Ideally, perfect mounting of the product within the inspection tool is given, lighting
conditions are the same across all inspections and types of defects are known in advance. With
AI-powered visual inspection, detection accuracy of defects increases while simultaneously
flexibility is enhanced, and deployment times decrease. The AI engine compares an image of
the tested part with thousands of reference images of parts with good quality to automatically
detect defects. Thereby, it does not depend on the exact position or lighting of the tested part. It
can learn to distinguish ‘good’ from ‘defect’ products by leveraging data from the existing quality
control process (e.g., cameras take images and the human decision (good/faulty) is associated
with these). This application is particularly interesting for SMEs who may not have a highly
standardized production line that outputs products in exactly the same position.

Economic impact:
AI enabled quality testing can lead to productivity increases of up to 50 percent. Efficiency and
speed are improved as the need for human input is lowered significantly. Improvements of up
to 90 percent in defect detection as compared to human inspection are feasible using deep-
learning-based systems. This reduces the costs associated with shipping defect products.
Generally, insights from AI-based quality testing can be used for root cause analysis to improve
the overall production process (McKinsey, April 2017). Depending on the starting point, the
impact can differ significantly: A light bulb producing SME employed 5 experienced workers
for quality control, which did not have to be replaced when they retired. An automotive supplier
needed to assure 100 percent quality in order not to lose its supplier contract with an OEM
which represented a critical part of their revenues (Expert interviews).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do require some degree of customization from

21
the company’s side. While several providers offer algorithms, these need to be integrated with
sensor hardware (e.g., cameras). The University of Amsterdam’s spin-off Scyfer, e.g., applies
deep neural networks to automate the visual inspection of steel surfaces leading to improved
performance especially in relation to complicated or rare defects. US-based Nanotronics
combines AI with 3D microscopy for defect detection down to the nanometer scale.

Capital requirements:
Machine-learning-based visual inspection costs less and is easier to implement than
conventional approaches (e.g., with pixel wise comparison) (McKinsey, 2017). With the
availability of open-source AI environments and relatively in-expensive hardware in terms
of cameras and powerful computers, even SMEs can increasingly rely on AI-based visual
inspection. Apart from the necessary hardware (e.g. high-speed camera) a main cost block is
implementation, which typically requires a data-scientist and a data engineer for several weeks
(e.g., an implementation of an MVP took 4 weeks when data was readily available (Expert
interviews).

Data availability:
This AI application mainly needs one type of data: Labelled quality data, e.g., images. These
need to be collected and labelled for the algorithm to learn: Images need to be annotated
with ‘good’ or ‘defect’. Ideally, this labelling process can be derived from the existing quality
inspection process (i.e., the cameras run in addition to human quality workers can learn from
their decisions). This can require several measurements (e.g., images from various perspectives
and different lighting conditions), to detect all kinds of quality issues.

Technological infrastructure:
To measure quality, some data on the product needs to be measured (images, sounds, x-ray,
etc.). This requires suitable sensors (e.g., high-speed cameras may be needed to achieve
necessary speed). The images then need to be transferred to the analytics solution via a
network (e.g., ethernet, Wifi). The AI algorithm can run on local hardware. However, for training,
large data sizes (thousands of labelled training images) and significant computation power
(especially graphical processing units (GPU)) is required. Therefore, training the algorithm needs
to be done in the cloud or through the solution providers capabilities.

Talent requirements:
There is no need for specific talent when the AI application is operational. For implementation,
data scientists and data engineers are needed, who are typically employed as external
consultants for several weeks.

22
Organizational and cultural challenges:
There are no organizational and cultural challenges

Job impact:
The introduction of this application could have a slightly negative impact on jobs. Depending on
prior quality control processes, human quality controllers may be substituted by the AI solution.

Environmental sustainability:
An indirect positive environmental effect is expected. A reduced number of returns could lead to
an indirect effect on environmental sustainability be saving resources spend on logistics (e.g.,
fuel, packaging).

Examples:

• A US-based manufacturer of valves and pumps achieved 100 percent accuracy in detecting
(non-)defective valves. (expert interviews)

2.8 Improving return on marketing invest by

optimizing marketing mix
Description:
Marketing budgets often represent a significant cost item, the impact of which is difficult to
measure. Marketing mix optimization aims to invest the marketing budget in the most efficient
way. Depending on the strategic objectives, this application can lead to higher sales with the
same marketing budget or save marketing costs without losing sales. To achieve this goal,
activities on different marketing channels (e.g., social media, online marketing, print) are
analyzed with respect to their resulting impact on sales. An AI algorithm looks at the marketing
actions (e.g., target group, time, investment) and the resulting sales (e.g., from tracking online
customers). It determines correlations between sets of inputs (e.g., an online ad for young
men in Berlin placed on a Friday evening) and resulting sales (e.g., sales of trousers to Berlin
addresses increasing in the following weeks). As a result, the value of 1 EUR spent can be
maximized by investing it in the most promising activity in the best channel. This is important
for SMEs who have a limited marketing. At the same time correlation between cause and effect
is easier to establish for SMEs as their marketing operations are not as complex as those from
larger companies.

Economic impact:

23
Optimizing the marketing mix can lead to a 10 to 20 percent increase in sales for a given
marketing budget (or a similar cost efficiency realization with constant sales) for SMEs. (Forbes,
February 2018)

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from
the company’s side. Providers of marketing mix modelling offer full outsourcing including the
collection of data, analysis and interpretation of results (e.g., Quantzig) They use several data
sources (e.g., company owned data, public sales statistics) to feed their model. As a result,
marketing investments are evaluated, leading to recommendations on which marketing actions
in which channels generate most sales.

Data availability:
The analysis is based on historical sales data and internal data (e.g., budget allocation). The
challenge lies in connecting a sale with previous marketing activities. This connection is easier
for social media and online channels, where customers can be relatively well targeted and data
is readily available. For example, a campaign on Facebook for 20- to 30-year-old women in a
specific region can be compared to sales in this group. For offline channels, such as a poster
campaign, a similar comparison is more challenging to link sales to the particular campaign.
In contrast to larger players, who often run several marketing activities in parallel, SMEs often
run a single campaign, which makes it easier to track effects. On the other hand, also for online
channels, data may be more difficult to acquire in the future, if intelligent tracking information
(e.g., third-party cookies) is increasingly blocked by browser providers such as Apple and
Firefox, who position themselves as protectors of user data (Medium, June 2019). Another
challenge lies in legal requirements for using personal data (see below).

Technological infrastructure:
There are no high computational requirements associated with the application. The software
runs as a software as a service (SaaS) application on any desktop computer and needs no
further technological infrastructure (apart from an internet connection). A basic requirement for
the implementation of this AI application is a customer relationship management (CRM) tool to
collect, categorize, and structure data in the individual sales channel.

Capital requirements:
Capital requirements are relatively low, depending on scope and current data availability.
Tracking the results of social media campaigns is included within respective platforms (e.g.,
Facebook). Website analytics are also available for free (e.g., Google Analytics) or a low fee per
month. Gathering data of offline campaigns is more challenging and costly, as it needs to be
manually collected and marketing actions need to be linked to respective sales. If marketing

24
expenditure and sales are already tracked with digital tools, the main cost factor lies in the
licensing cost of the AI solution. Otherwise, this data first needs to be collected, which might
involve a customer relationship management (CRM) software.

Talent requirements:
Off-the shelf solutions can be implemented without the need for specialized AI talent. However,
some knowledge of digital marketing might be required, to implement recommendations.

Organizational and cultural challenges:

The organization impact is limited if the changes to the marketing mix are relatively small. As
these changes grow, more significant changes in personnel allocation and potentially retraining
might be required (e.g., for digital marketing activities).

Job impact:
This application has no impact on jobs.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• A consumer goods company improved the marketing budget allocation, optimized the
marketing mix and introduced “one currency” to measure and compare marketing return on
investment (ROI). This yielded an increase of around 15 percent in sales revenues (expert
interviews).
• Another consumer goods company reduced its marketing spend by 15 percent (expert
interviews, same).

2.9 Improving reliability at lower cost by using

predictive maintenance
Description:
Predictive maintenance aims at improving asset productivity by using data to anticipate
machine breakdowns. This can lower maintenance cost as maintenance can be scheduled
and performed before serious and thus more costly failures occur. Furthermore, downtimes
are reduced increasing operating equipment efficiency. To predict failures, sensor data (e.g.,
sounds, temperature, vibrations, visual data) from machinery and equipment is interpreted

25
to detect abnormalities (e.g., unusual sounds, temperature, vibrations). Data sources can be
manifold and depend on the specific application. For example, in engines, gear boxes, or air
conditioning, the analysis of sound can detect an anomaly in device operation. In switches,
machines, and robots, vibrations can be measured and used to identify errors. AI-based
algorithms are capable of recognizing patterns in the data that hint at an (upcoming) error and/
or failure of the system. Thus, they are able to predict breakdowns even before they happen and
allow to prevent them by performing timely maintenance.

Economic impact:
Depending on the exact circumstances, inspection costs may be reduced by 10 to 25 percent,
and an overall reduction of 5 to 10 percent of annual maintenance costs is possible. Depending
on the starting point (i.e., if machines are running at high utilization and no backups are
available to make production more flexible), availability can sometimes increase by more than
20 percent (McKinsey, April 2017).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side. Tools are available at equipment vendors (e.g., IBM, Siemens, GE, ABB) as well
as software firms (IBM, Microsoft) and dedicated companies (e.g., Predikto, Konux). These
are especially feasible for SMEs as they allow for easy integration of predictive maintenance
in existing workflows. Integration of several providers may pose a challenge if the data is not
directly compatible.

Data availability:
Two types of data are needed or implementing predictive maintenance solutions. First, a data
set is required as a basis for comparison with the current measurements. This may be obtained
by measurements over time, including the labelling of respective incidents (which is cost- and
time-intensive) or bought from, e.g., the equipment manufacturer. One very simple method
would be to measure the power consumption of a motor during normal operating conditions
and set this as a baseline. Second, current (sometimes live) data from machines and/or
equipment is needed to detect anomalies. This may be available because it is already measured
(e.g., temperature of cooling liquid is often monitored for safety reasons) or may need to be
gathered through additional sensors (e.g., the vibration pattern of a legacy hot press is usually
not measured). As the comparison with historical data is usually done on a central piece of
hardware or in the cloud, the data from different sensors needs to be transmitted to the analysis
hub via some sort of connection (cable, Wi-Fi, etc.).

Technological infrastructure:
In the simplest case, the machine provides a digital measurement (e.g., the power consumption

26
of a motor), which can directly be used for analysis (if power consumption goes up under
normal working conditions, there might be a problem). Then, only a network connection and
computation power are necessary, which may be local or cloud based. In other cases, i.e., when
machine data is not sufficient, additional sensors need to be installed (e.g., to measure unusual
vibrations or sounds) and also connected to the network or cloud system.

Capital requirements:
Capital requirements depend on the starting point, scope, and ambition. If a company has one
type of machine that already measures relevant data (e.g., power consumption of a motor)
and ideally already shares this data over a firm network, the investment concerns only the AI
solution itself. In contrast, if data needs to be measured, digitized, and connected via a network
built for this purpose, costs are significantly higher.

Talent requirements:
Simple solutions can be installed by external providers. Their use does not require additional
talent. Larger and more complex production environments may require experts for the
integration of various equipment vendors and may also require continuous support by experts.
The more customized and variable a production process, the more likely it is to need AI experts
to continuously develop the predictive maintenance solution.

Organizational and cultural challenges:

No organizational and cultural challenges are expected.

Job impact:
The introduction of this application could have a slightly negative impact on jobs. As
maintenance work is reduced, less maintenance staff is necessary. However, some additional
work is created to maintain and supervise the predictive maintenance system, thus up- and
reskilling of some job profiles might be necessary. Furthermore, maintenance work will be
easier to plan and can be performed largely during normal working hours (as opposed to night
shifts on alert).

Environmental sustainability:
An indirect positive environmental effect is expected. It results from reduced waste production
(e.g., a plastic producer will have less faulty output due to machine failure) and longer lifetime,
resulting in fewer new machines needed (i.e. fewer resources to build them).

Examples:

• In a North American copper mine, 12 percent annual savings in maintenance cost (of a total

27
 of more than EUR 4.5 million) were achieved through optimized maintenance strategies.
• A chemicals company reduced maintenance costs related to pump breakdowns by 70
percent.
• A German semiconductor manufacturer achieved 90 percent accuracy in predicting
downtimes four weeks in advance, while 10 percent maintenance capacity freed up and
production waste due to faulty machinery was reduced by 40 percent.

28
2.10 Preventing and detecting cyberattacks by
analyzing threat data and network traffic
Description:
Cyberattacks pose a serious risk to businesses. For example, a leading Japanese manufacturer
suffered an attack that led to a partial shutdown of its production lines for three days, dropping
production output by 50 percent (Capgemini, July 2019). Traditional (non-AI) cybersecurity
systems detect threats based on rules which are derived from historical threat data. This does
not protect against new viruses and malware (in the same way last year’s flu immunization does
not protect against this year’s flu strain). AI can be trained with pattern recognition in software
to detect viruses and malware attacks before they enter the system and isolate them. The
algorithm learns from data structure in existing viruses and malware, the behavior of software
within the system, and network traffic (e.g., which data are going into and out of the network) to
differentiate between good and malicious activity. As SMEs often do not have the cybersecurity
resources and experts to identify, evaluate, and act on cybersecurity threats, AI can be crucial
for them to mitigate risks from the continuously increasing threat landscape (Klahr et al., April
2017).

Economic impact:
The reduction in costs to detect and respond to breaches ranges from 1 to 15 percent, with
an average of 12 percent, according to a survey across 850 organizations from 10 different
countries mainly in Europe (Capgemini, July 2019). With AI, the time taken to detect threats
and breaches can be reduced by up to 12 percent. First, this means reactions can be faster
and prevent further damage. Second, it also reduces labor time. AI also reduces the time taken
to remediate a breach or implement patches in response to an attack by around 12 percent
(Capgemini, July 2019). For SMEs who currently invest little or nothing in cybersecurity because
it is difficult to attract IT talent (Smallbusiness.co.uk., March 2019), this AI application helps
reduce their risk exposure by substituting a cybersecurity department.

Availability of off-the-shelf solutions:

Many companies offer services ranging from supporting businesses’ security analysts with
threat analytics (e.g., Anomali, LogRythm) up to a full-service offering, including reaction (e.g.,
Cyberreason, CrowdStrike, eSentire). Further solutions are expected to be made available soon.
For example, GE developed Digital Ghost to detect cyberattacks by using digital twins of the
machine control systems, and Siemens’ Industry Anomaly Detection identifies anomalies in
industrial networks.

29
Capital requirements:
The costs of off-the-shelf solutions are often based on the number of endpoints (e.g., laptops,
smartphones). A detection solution may start at EUR 50 to 70 per endpoint per year, while a
full-service solution (from prevention and detection to response) costs around EUR 200 per
endpoint per year (CrowdStrike, n.d.).

Data availability:
For this AI application to work, it needs to learn from existing cybersecurity data (e.g., viruses,
malware) as well as network behavior. Data on malware and viruses is gathered by the off-the-
shelf providers themselves or sourced externally from companies specializing in this business.
This data is then continuously fed into the AI tool to detect new threats. Network behavior is
partly generic (e.g., endpoints like computers communicating within the local network and with
internet addresses) and can be pretrained by service providers but partly needs to be trained
individually for each business (e.g., a proprietary software might produce unusual but harmless
traffic).

Technological infrastructure:
As off-the-shelf-solutions exist that can be run on any computer or in the cloud, there are no
specific requirements with regard to technological infrastructure.

Talent requirements:
Some off-the-shelf solutions allow the full outsourcing of cybersecurity to the service provider.
They also manage integration either on the premises or in the cloud. If only the threat analysis is
outsourced, IT security specialists will need to continue to respond to the threat or attack in an
adequate manner.

Organizational and cultural challenges:

There are no organizational and cultural challenges expected.

Job impact:
The potential impact on jobs is unclear. On the one hand, AI can help to reduce cybersecurity-
related work, leading to lower labor costs. On the other hand, many SMEs do not have dedicated
cybersecurity employees, so there is no impact on labor, only on risk.

Environmental sustainability:
No direct or indirect environmental effects are expected.

30
Examples:

• ED&F Man, a British commodities trading company, integrated an AI based cyber-security. It

cut incident investigation time from hours to minutes and reduced priority alerts from 800
to five per month (Vectra.ai, n.d.)

• A global aerospace manufacturer streamlined its endpoint security process and automated
threat detection with AI powered cyber-security (Cyberreason, n.d.)

2.11 Enhancing the efficiency of legal document

generation and review by using natural
language processing and automation

Description:
Natural language processing can be employed for text mining purposes (i.e., extracting
information from written resources) and for the review and creation of contracts. It enables the
management of large pools of documents, including searching across documents, workflow
automation, and automatic language detection. Modern solutions apply machine learning
algorithms to automatically extract and analyze information, transfer knowledge across
contracts, and highlight those requiring review. Documents can be imported in a variety of
formats and are automatically converted into machine-readable form. Machine learning models
then identify specified concepts and clauses. On-screen editing and workflow tools make it easy
to refine results, collaborate in real-time, and view changes made by different reviewers. Search
and analysis tools facilitate the identification of issues and trends and can compare these
across documents. Document management tools can highlight the total number of documents
and documents reviewed to easily create project summaries and charts (Kira, n.d. c).

Overall, the combination of AI applications and human reviewers yields more precise results
than either of them separately (Clifford Chance, 2017). Without technology, human reviewers
find only up to 65 percent of instances, and 75 to 80 percent with the addition of traditional
technological assistance (which is not AI-based) (Grossman & Cormack, 2011).

Economic impact:
The economic impact can differ depending on the provider and context. Contract review and
analysis can be completed in 20 to 90 percent less time (20 to 40 percent less for first-time
users, up to 90 percent less for more experienced users) (Kira, n.d. c).

31
Availability of off-the-shelf solutions:
There are off-the-shelf solutions available for a part of the application. For example, the
Canadian provider Kira built a proprietary machine learning algorithm to provide more than
900 built-in provision models to cover a variety of use cases (e.g., due diligence, M&A deal
points, credit agreements, general commercial, non-disclosure agreements, real estate).
Simultaneously, the algorithm can learn from examples based on sample contracts uploaded
by the user that highlight the clauses they want the tool to learn. Based on monitored accuracy
levels of the learned models, the user can decide when to allow Kira to automatically identify the
relevant clauses.

Data availability:
Natural language processing requires access to the documents to be reviewed. Files in any
format can be used, including old scans.

Technological infrastructure:
Since most of these applications are off-the-shelf-solutions that can be run on any desktop PC
or in the cloud, no specific technological infrastructure is required.

Capital requirements:
Capital requirements are low and limited to the subscription fee of the software suite used.
Pricing is based on volume of usage and is sold in annual license contracts.

Talent requirements:
When using off-the-shelf solutions, SMEs do not need to hire specialized talent. Employees
should be trained to use the software. Thanks to user-friendly interfaces and the availability of
support staff/portals, this is not usually time-intensive.

Organizational and cultural challenges:

The primary obstacle to overcome in this context is acceptance from legal departments. They
may not fully trust software applications to thoroughly understand all the nuances of complex
legal documents and thus, companies may be initially reluctant to rely on these applications in
the context of their daily work. However, where such applications are employed safely as tools
for eliminating laborious manual tasks, they can greatly improve employee experience, thus
enabling greater adoption and acceptance.

Job impact:
There might be a slightly negative impact on jobs, as these applications can enhance the speed
and efficiency of the document review process. However, current applications can only support
processes, not replace legal staff completely.

32
Environmental sustainability:
An indirect positive environmental effect is expected, as legal documents can be stored, edited,
and shared online, reducing the need to print them.

Examples:

• A 275-attorney US law firm used Kira in its M&A practice to streamline and accelerate the
document review process. The company was tasked with reviewing over 500 documents
(including NDAs, employer agreements) within five working days. Using Kira, the firm
completed the project early, saving approximately 48 percent of the person hours required
for the project (Kira, n.d. d).

33
2.12 Optimizing prices by identifying margin
improvement and price perception
opportunities

Description:
Companies may improve their margins by employing price optimization techniques that
automatically counter changes in price-determining factors, such as price competition or
customer price elasticity. These applications draw on algorithmic repricing rules, which are
informed by relevant price competition for strategically relevant stock-keeping units (SKUs).
Hence, if there are price movements observed among competitors for products that drive
margins, the software allows to respond and keep margins stable by shifting them to other
SKUs or following the competitors’ moves. The impact and quality of these pricing strategies
depends on external information. Thus, they are more commonly employed in B2C than in
B2B enterprises, where the public availability of price data and product variety allows for more
efficient applications. The application is particularly relevant for SMEs as there tends to be less
price information regarding the products they sell, which means they have greater potential to
capture additional margin.

Economic impact:
The applications are expected to increase revenues by 3 to 5 percent, which ultimately
translates into margin increases, the impact being a function of the SKUs affected by the pricing
strategy. The economic potential for SMEs is probably greater than for large retailers, as they
are typically exposed to lower demand elasticity, driven by more fragmented competition and
less available price information. (expert interviews)

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side. Most solutions are subscription based and allow for a complete outsourcing
of price optimization activities. Well-known vendors include Periscope, Competera, and Wise
Athena, among others.

Data availability:
Data sourcing is especially difficult for SMEs as price optimization is driven by regional
competition, and location-specific data is difficult to obtain. Physical prices need to be collected
manually, e.g., it is common for large retailers for example to hire people who buy and scan

34
representative product baskets from their competitors to gain price information (e.g., a “test
family” is paid to do their daily purchases, scan receipts, and play them back to the retailer that
engaged them; alternatively, this can also be outsourced). On the other hand, a lack of local
competition might restrict the volume of available data.

Technological infrastructure:
The computational requirements for running price optimization algorithms vary dramatically
with the types of products sold, the availability of competitors’ price information, and SKUs of
interest. For example, a large retailer with countrywide operations may need to run the software
with dozens of GB for every request, at which point cloud access would be necessary. For SMEs,
local computation or the use of software on a local device would be fully sufficient. If data
volumes allow for the use of off-the-shelf solutions that can be run on any computer or in the
cloud, there are no specific requirements with regard to technological infrastructure.

Capital requirements:
They depend primarily on the technological infrastructure used. Cloud access is more expensive
than computation with off-the-shelf SaaS solutions. Additionally, the frequency of computation
is a key price determinant (e.g., cloud calls may be paid for by use in contrast to on-premise
computation). Companies with seasonal product mixes and frequent marketing campaigns
and promotions will more often require a recomputation of their pricing strategies and thus be
naturally exposed to higher capital requirements.

Talent requirements:
No additional talent is required when an off-the-shelf solution is adopted, and maintenance
is completely outsourced. If proprietary algorithms are developed and executed in the cloud,
then requirements are higher (e.g., would require developer with data science, econometrics
background, potentially cloud certification).

Organizational and cultural challenges:

No organizational challenges are expected, as tool has no structural impact on any business
function, but rather serves as a support for pricing decisions.

Job impact:
The introduction of this application could have a slightly positive impact on jobs. Additional
labor might be required if the process is not outsourced. Growth in revenue may ultimately
cause an increase in labor demand.

35
Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Find Me A Gift, a British e-commerce retailer, achieved a 25 percent sales increase after
changing prices based on algorithmic predictions, partnering with Competera to receive
competitive data. (Competera a)
• Foxtrot, an Eastern European consumer electronics retailer, achieved a 15 percent sales
increase after employing the price optimization software of Competera, using databases on
nonlinear interconnections. (Competera b)
• Consumer-packaged goods with large SKU bases (e.g., Barilla) may achieve 2 to 6 percent
contributional margins, leveraging Wise Athena’s platform to predict prices for and demand
of SKUs during promotional cycles. (Wise Athena)

36
2.13 Optimizing production by forecasting demand
Description:
Predicting demand enables producers to make the right product in the right quantity at the right
time. This can increase overall sales as demand can be met during peak times that would not
be met without proper prediction. Furthermore, inventory levels can be reduced when inventory
is adapted to production needs. While traditional demand forecasting models are mainly
based on historical sales, AI can help to uncover non-linear patterns (e.g., dependencies on
weather). The AI solution tries to find correlations between combinations of input data (e.g.,
weather, demographics) and resulting sales. It might, e.g., find that good weather on certain
days combined with holidays in a certain region leads to increased sales. Demand forecasting
is especially important for SMEs to remain competitive as their production often does not profit
from the same scale effect that larger companies leverage.

Economic impact:
AI-based approaches to forecasting can reduce forecasting errors by 30 to 50 percent in
some settings. Furthermore, lost sales due to product unavailability can be lowered by up to
65 percent. Costs related to transport and ware-housing, and supply chain administration are
expected to decrease by 5 to 10 percent and 25 to 40 percent, respectively. Due to AI, overall
inventory reductions of 20 to 50 percent are feasible. (McKinsey, April 2017)

Availability of off-the-shelf solutions:

There are off-the-shelf solutions ((e.g. Relex, Blue Yonder, Lokad) available which do not require
further customization from the company’s side.These have to be integrated into business data
systems by technical consulting staff.

Capital requirements:
There are two types of costs associated with this AI application: The costs of the software
itself (roughly EUR 20,000 to 40,000) and the cost of the implementation, ranging roughly from
EUR 20,000 to 50,000 (Demand-Planning.com, n.d., Silvon.com n.d.). The implementation effort
depends on the sophistication and complexity of existing data management systems. For
example, if there is one consolidated CRM system with an accessible data interface, integration
will be relatively straightforward. If more internal systems of the company (e.g., ERP system) are
connected set-up will become more complicated as additional interfaces will be needed.

Data availability:
To forecast demand, two types of data are needed: sales data (e.g., orders fulfilled, orders
missed) and potentially influencing parameters (e.g., location, weather, demographics). Sales
data are typically available within customer relationship management systems (CRM). Other

37
data (e.g., weather) is typically either publicly available or can be bought from specialized data
providers.

Technological infrastructure:
There are no high computational requirements associated with the application. The software
runs as a software as a service (SaaS) application on any desktop computer and requires no
further technological infrastructure (apart from an internet connection).

Talent requirements:
Implementation of off-the-shelf solutions is feasible without any specific in-house analytics. The
application can run without the need for AI experts. However, data scientists and engineers can
help to reduce implementation efforts and continuously adapt the solution, to new scenarios.

Organizational and cultural challenges:

No major changes are required when existing demand forecasting schemes are enhanced
with this solution. Even if the organization starts the demand planning from scratch, no major
changes are required. However, depending on the variability of forecasts (e.g., a large change in
quantity), a closer integration with procurement and production planning may be necessary to
reap the full benefits of this AI application.

Job impact:
The introduction of this application could have a slightly positive impact on jobs. If proper
demand planning results in increased sales, additional jobs might be created across business
functions (e.g., in production, marketing, sales, logistics)

Environmental sustainability:
An indirect positive environmental effect is expected. If left-over production and inventory can
be reduced, this saves resources, leading to an indirectly positive environmental impact.

Examples:

• A Brazilian personal care producer employed demand forecasting and reported a 27 percent
reduction in shortage (sales, that cannot be fulfilled) and 22 percent reduction in inventory.
(expert interviews)
• A consumer packaged goods company improved forecast accuracy to 95 percent three
months into the future starting from 83 percent 10 days into the future. (expert interviews)
• OTTO, a German retail player, introduced an AI system to predict sales of the next 30 days
with 90 percent accuracy. (The Economist, April 2017)

38
2.14 Enhancing the productivity of the pick-and-
place process in warehouses by using smart
glasses with augmented reality
Description:
Smart glasses comprise computer technology to allow the wearer to view information. Smart
glasses facilitate the pick-and-place process in warehouses by providing information about the
object and quantity and guiding the operator to the respective location. As a hands-free solution
that is permanently connected to the warehouse management system and incorporates
microphones and speakers, smart glasses can render traditional order picking techniques with
maps (e.g., handheld devices or paper-based maps which tie up a picker’s hands) or voice
guidance (e.g., headphones which rely on the picker to remember the instructions) obsolete.
Smart glasses use vision picking (i.e., scanning barcodes by looking at the object) to identify
printed codes attached to a pallet or shelf space item in a warehouse, overlay graphics on
top of these codes, and indicate the volume to collect as well as direct the picker to the next
item to be collected (vuzix.eu, n.d.). Smart glasses can be integrated with any existing back-
end system that tracks orders and supplies layout information (e.g., warehouse management
system (WMS), transportation management software (TMS), enterprise resource planning
(ERP)) about the warehouse), facilitating relatively quick implementation. While some smart
glasses have a micro heads-up display (HUD) on the side of the vision field, other smart
glasses use augmented reality to overlay translucent visuals in the middle of the vision field
(Michel, November 2018). Augmented-reality smart glasses have integrated sensors (e.g.,
cameras, accelerometers) and AI-powered visual recognition techniques to grasp the physical
characteristics of the surrounding and detect walls, floors, and surfaces. Paired with a built-in
GPS, smart glasses are aware of the operator’s location in the warehouse and direct operators
to the correct location of the item or shelf they are looking for (Michel, November 2018). In
addition to facilitating pick-and-place, smart glasses have a video recording/feedback function
for prompt error detection, which can be used for training purposes, e.g., to highlight operational
errors or safety hazards (vuzix.eu, n.d.). Finally, smart glasses may enhance the productivity of
warehouse logistics and thereby decrease labor costs.

Economic impact:
Using smart glasses may increase productivity by 25 to 30 percent compared to paper-based or
radio frequency (RF) scanning alternatives, yielding a payback period of less than 0.6 years and
reducing errors by up to 60 percent compared to paper-based picking (smartpick.be, n.d.).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the

39
company’s side (e.g., Smartpick, Vuzix, HoloLens). The solutions offer various configurations
and different technical capabilities (e.g., lensless frames, impact-resistant glasses for eye
protection, gigabyte storage capacity, superior camera resolution). Moreover, smart glasses are
likely to be sold as part of a package of warehouse management system vendors in the future
(postscape.com, n.d.).

Data availability:
Three types of data are required: (1) data on the warehouse layout and the position of individual
products within the warehouse (e.g., alley number, shelf number), which can usually be sourced
from warehouse management software (e.g., from ERP, TMS, WMS), (2) data on the position
and orientation of the picker (e.g., truck located on alley 2, looking north), which is usually
gathered by the smart glasses, e.g., through a compass or motion sensors, and (3) data on the
products that should be picked and combined in one shipment and the location(s) for packing
of mentioned goods, which is usually available from the warehouse or customer management
software. (Williams, June 2018)

Technological infrastructure:
Off-the-shelf solutions run on any computer but require access and appropriate interfaces
(APIs) to a back-end system (e.g., ERP, TMS, WMS). Furthermore, this application requires a
wireless internet connection, a printer to print the barcodes for scanning, and smart glasses
(smartpick.be, n.d.). Depending on the configurations of the warehouse, wearable Bluetooth ring
scanners are recommended to scan items not within viewing height.

Capital requirements:
Depending on the provider and features mentioned above, the cost of higher-end smart glasses
vary between EUR 900 to EUR 3,000 (vuzix.eu).

Talent availability:
This application does not require any special human capital or talent. The software suites are
easy to use with intuitive interfaces. According to some providers (e.g., Smartpick), 15 minutes
of training are required until the picking team is ready to begin its work.

Organizational and cultural challenges:

Organizational challenges are expected as some workers may voice concern or unease about
being subject to continuous monitoring.

Job impact:
The introduction of this application could have a slightly negative impact on jobs. As the pick-
and-place process is simplified, productivity is enhanced because the process takes less time.

40
Therefore, workers’ capacity is freed up while demand is not necessarily increasing.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Bakery warehouses in Belgium use smart glasses to distribute dozens of types of bread
to hundred of stores. This reduced errors by 60 percent while increasing productivity by 25
percent (postscapes.com, n.d.).
• A Belgian sector-focused night distributor combined smart glasses with a Bluetooth ring
scanner (for more scan-intensive duties) to quickly sort small packages. The operator scans
a package and the application indicates the bin to place the package in. This process takes
less than three seconds and increased productivity by 30 percent (smartpick.be, n.d.).
• DHL, a German contract logistics provider, piloted the use of HUD smart glasses on unit
picking to a card and incorporated it as part of a multimodal solution in which operators
provided voice confirmations through the smart glasses and used ring scanners. After
the successful pilot, the company implemented the glasses as a standard in warehouse
operations, which has improved the average productivity by 15 percent (logistics.dhl, 2019).

2.15 Enhancing maintenance by using smart glasses

Description:
When a machine stops working, this can lead to significant production downtime when no
experienced maintenance technician is available to fix the error quickly. This is particularly
challenging for SMEs with a limited number of technicians. Smart glasses can help by providing
a direct link to remote technicians (e.g., at the machine manufacturer). They combine sensors
(e.g., camera, microphone, bar code scanner) and augmented reality. This enables an on-site
technician to wear the smart glasses and perform the required maintenance work. Via network
connection (e.g., WiFi, Bluetooth), a remote expert monitors the on-site technician’s actions
and guides the repair through voice instructions and augmented reality. To do so, the camera
films the technician’s field of view. Through image recognition, the environment is screened
(e.g., machine parts) and points out relevant information (e.g., which parts to remove, which
screws to unscrew). Furthermore, the sensors may detect maintenance issues that the on-site
technician is not aware of (e.g., by picking up a noise that points to another part that needs to
be replaced) (Forger, August 2018).

Economic impact:
Smart glasses can be used to reduce machine downtimes, increase the rate of first-time fixes
to reduce maintenance costs, and improve the overall effectiveness of production equipment.

41
High impact is expected in cases of high utilization or just-in-time production with fines for not
delivering promised output, and if the time from the arrival of the maintenance worker on-site to
the start of work can be significantly shortened.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side (e.g., Vuzix). These solutions include hardware (smart glasses) and additional
services (e.g., video calling). They are ready to use and require no further customization.
Additionally, some smart glasses are linked to a 24/7 service hub (e.g., Honeywell’s Techsight).
Smart glasses can also be customized by combining hardware (e.g., Epson, Realware) with
applications such as video calling (e.g., Oculavis, RemoteEye).

Data availability:
No data is needed for this application, except for the data that is picked up by the sensors when
the smart glasses are in use. aFor remote maintenance support, technicians need to either have
the relevant knowledge or access to it (e.g., user manuals).

Technological infrastructure:
Apart from the smart glasses itself, a high-speed internet connection is required to transmit the
video.

Capital requirements:
Costs are typically split between the hardware (one-off costs) and additional services (e.g.,
video calling). Prices for smart glasses start at a few hundred euros (e.g., EUR 750 excluding
VAT (Epson, n.d.)). A video calling app that transmits video calls over the internet, for example,
costs about EUR 50 per user per month (Vuzix, n.d.).

Talent requirements:
This application does not require any special human capital or talent. The smart glasses are
easy to use with intuitive interfaces. They do not require significant training.

Organizational and cultural challenges:

When smart glasses are introduced, maintenance workers might need some time to get
used to the devices, e.g., wearing them or remember charging the batteries. However, no
large challenges are expected as smart glasses significantly simplify maintenance work.
Nevertheless, it may be challenging for some maintenance workers to be closely monitored and
receive detailed instructions as opposed to working on their own and deciding for themselves
how to fix a problem.

42
Job impact:
The introduction of this application could have a slightly negative impact on jobs, as there is no
need for specialized technicians on-site, and repairs can be carried out more efficiently.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• German carmaker BMW introduced smart glasses at US dealerships to reduce repair times
and improve customer satisfaction (BMW, June 2019).
• German machine manufacturer MAN Energy Solutions is currently testing its in-house
developed assisted-reality application PrimeServ EyeTech (Sampson, December 2020).

2.16 Automating invoice processing by visually

recognizing invoice data
Description:
Managing accounts payable requires processing invoices from arrival of the invoice till the
fulfilled payment. The invoice first needs to be preprocessed by inputting relevant data (e.g.,
amount payable, reference number, bank account details) into the enterprise resource planning
(ERP) system (e.g., SAP, MS dynamics). It is then validated (e.g., compared with the respective
order) and finally a payment of a certain amount has to be initiated at a certain time. A large
part of the work in processing payable invoices is manual data entry, i.e., reading data off an
invoice and inputting them into an ERP system. This is a tedious task, which is prone to small
mistakes (e.g., transposed digits). AI can take over this task and can ensure that data is almost
always correctly processed (e.g., more than 99 percent accuracy with Itemize or Workfusion
(Itemize, n.d.)). To process invoices automatically, the invoices (e.g., printed, pdf) are scanned
and uploaded into the tool. It then extracts the relevant information (e.g., amount, invoice
number, date) using image recognition and natural language processing. The data is then
fed into the ERP system (e.g., SAP, MS dynamics) via an interface (API). From there, further
processing can be initiated, including payment of the invoice. Like other back-office tasks,
invoice processing is not a core competency of manufacturing companies and therefore one
that is often not optimized. Automating helps SMEs to remain competitive and also be more
flexible in responding to peak periods, which is often more difficult for smaller companies with
limited staff.

43
Economic impact:
This AI application increases back-office efficiency by reducing process time and labor. The
process of invoice data entry can be fully eliminated. Assuming an average data entry time of 3
minutes per invoice and 200 invoices per week, this AI application would save ten hours every
week.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from
the company’s side (e.g., by Rossum, Parashift, Tipalt, Itemize). They can process any human
readable invoice as soon as it is digitized (e.g., uploaded from a scanner). They output data into
standard ERP systems (e.g., SAP, MS Dynamics) through an interface (API). The integration is
straightforward and supported by the vendors.
Capital requirements:
There are different pricing models for this AI solution, mostly customized to the respective
business. Generally, the cost of this AI solution depends mainly on the number of invoices to
be processed (e.g., EUR 0.5 per invoice up to 2,500 invoices per month and EUR 0.2 for any
additional invoice (Hypathos, n.d.)). Additional functions (e.g., approvals routing) also increase
the price.

Data availability:
There is only one data source needed for this AI application: invoices in any format (e.g., printed
invoices, pdfs) can be used. Paper invoices have to be digitized with a scanner, for the AI
application to be able to process them.

Technological infrastructure:
As off-the-shelf-solutions exist that can be run on any desktop PC or in the cloud, no specific
requirements with regard to technological infrastructure exist. Only a scanner is required to
digitalize invoices.

Talent requirements:
There are no AI talent requirements as off-the-shelf solutions for this AI application are
available. Some expertise is required to set up the interface (API) between the AI solution and
the ERP system. However, this is typically done by external consultants.

Organizational and cultural challenges:

When the AI solution is first introduced, it may be challenging to convince financial staff that the
data processing always yields the right result. This can be mitigated by introducing a validation
step for humans to verify the data readout in the initial phase of integration.

44
Job impact:
The introduction of this application could have a slightly negative impact on jobs, as this AI
application automates a process that is currently performed by a human. The saved time can be
invested in other activities (e.g., increased workload from financial processes if the company is
growing).

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• A US-based beverage company reduced invoice data entry to SAP from 3 minutes to a few
seconds. (Rossum, n.d.)
• A Czech online retailer, DoDo, reduced document processing time by 75 percent (Rossum,
n.d.)

2.17 Enhancing the productivity of the pick-and-

place process in warehouses by using smart
glasses with augmented reality
Description:
Smart glasses comprise computer technology to allow the wearer to view information. Smart
glasses facilitate the pick-and-place process in warehouses by providing information about the
object and quantity and guiding the operator to the respective location. As a hands-free solution
that is permanently connected to the warehouse management system and incorporates
microphones and speakers, smart glasses can render traditional order picking techniques with
maps (e.g., handheld devices or paper-based maps which tie up a picker’s hands) or voice
guidance (e.g., headphones which rely on the picker to remember the instructions) obsolete.
Smart glasses use vision picking (i.e., scanning barcodes by looking at the object) to identify
printed codes attached to a pallet or shelf space item in a warehouse, overlay graphics on
top of these codes, and indicate the volume to collect as well as direct the picker to the next
item to be collected (vuzix.eu, n.d.). Smart glasses can be integrated with any existing back-
end system that tracks orders and supplies layout information (e.g., warehouse management
system (WMS), transportation management software (TMS), enterprise resource planning
(ERP)) about the warehouse), facilitating relatively quick implementation. While some smart
glasses have a micro heads-up display (HUD) on the side of the vision field, other smart
glasses use augmented reality to overlay translucent visuals in the middle of the vision field
(Michel, November 2018). Augmented-reality smart glasses have integrated sensors (e.g.,

45
cameras, accelerometers) and AI-powered visual recognition techniques to grasp the physical
characteristics of the surrounding and detect walls, floors, and surfaces. Paired with a built-in
GPS, smart glasses are aware of the operator’s location in the warehouse and direct operators
to the correct location of the item or shelf they are looking for (Michel, November 2018). In
addition to facilitating pick-and-place, smart glasses have a video recording/feedback function
for prompt error detection, which can be used for training purposes, e.g., to highlight operational
errors or safety hazards (vuzix.eu, n.d.). Finally, smart glasses may enhance the productivity of
warehouse logistics and thereby decrease labor costs.

Economic impact:
Using smart glasses may increase productivity by 25 to 30 percent compared to paper-based or
radio frequency (RF) scanning alternatives, yielding a payback period of less than 0.6 years and
reducing errors by up to 60 percent compared to paper-based picking (smartpick.be, n.d.).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side (e.g., Smartpick, Vuzix, HoloLens). The solutions offer various configurations
and different technical capabilities (e.g., lensless frames, impact-resistant glasses for eye
protection, gigabyte storage capacity, superior camera resolution). Moreover, smart glasses are
likely to be sold as part of a package of warehouse management system vendors in the future
(postscape.com, n.d.).

Data availability:
Three types of data are required: (1) data on the warehouse layout and the position of individual
products within the warehouse (e.g., alley number, shelf number), which can usually be sourced
from warehouse management software (e.g., from ERP, TMS, WMS), (2) data on the position
and orientation of the picker (e.g., truck located on alley 2, looking north), which is usually
gathered by the smart glasses, e.g., through a compass or motion sensors, and (3) data on the
products that should be picked and combined in one shipment and the location(s) for packing
of mentioned goods, which is usually available from the warehouse or customer management
software. (Williams, June 2018)

Technological infrastructure:
Off-the-shelf solutions run on any computer but require access and appropriate interfaces
(APIs) to a back-end system (e.g., ERP, TMS, WMS). Furthermore, this application requires a
wireless internet connection, a printer to print the barcodes for scanning, and smart glasses
(smartpick.be, n.d.). Depending on the configurations of the warehouse, wearable Bluetooth ring
scanners are recommended to scan items not within viewing height.

46
Capital requirements:
Depending on the provider and features mentioned above, the cost of higher-end smart glasses
vary between EUR 900 to EUR 3,000 (vuzix.eu).

Talent availability:
This application does not require any special human capital or talent. The software suites are
easy to use with intuitive interfaces. According to some providers (e.g., Smartpick), 15 minutes
of training are required until the picking team is ready to begin its work.

Organizational and cultural challenges:

Organizational challenges are expected as some workers may voice concern or unease about
being subject to continuous monitoring.

Job impact:
The introduction of this application could have a slightly negative impact on jobs. As the pick-
and-place process is simplified, productivity is enhanced because the process takes less time.
Therefore, workers’ capacity is freed up while demand is not necessarily increasing.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Bakery warehouses in Belgium use smart glasses to distribute dozens of types of bread
to hundred of stores. This reduced errors by 60 percent while increasing productivity by 25
percent (postscapes.com, n.d.).
• A Belgian sector-focused night distributor combined smart glasses with a Bluetooth ring
scanner (for more scan-intensive duties) to quickly sort small packages. The operator scans
a package and the application indicates the bin to place the package in. This process takes
less than three seconds and increased productivity by 30 percent (smartpick.be, n.d.).
• DHL, a German contract logistics provider, piloted the use of HUD smart glasses on unit
picking to a card and incorporated it as part of a multimodal solution in which operators
provided voice confirmations through the smart glasses and used ring scanners. After
the successful pilot, the company implemented the glasses as a standard in warehouse
operations, which has improved the average productivity by 15 percent (logistics.dhl, 2019).

47
2.18 Increasing workers’ safety by using machine
vision
Description:
Safety at work is a top priority for managing human and financial risks. Yet, many SMEs
operate with scarce expert safety resources. In some environments, cameras are installed for
security and other reasons. The analysis of the available footage (images and videos which are
aggregated as part of daily operations) may lead to an overall increase in workers’ safety. Using
AI-based image and pattern recognition techniques to detect safety hazards (e.g., workers not
wearing hard hats) and noncompliance with safety standards from the visual data may help
prevent accidents.

Economic impact:
According to the Suffolk Construction Company and Smartvid.io, a contractor would save
an average of EUR 30,000 per each prevented accident (including compensation payments
for workers and economic loss caused by worker’s absence after the incident; US estimates
(Boston Globe, October 2019)). The application may reduce incidents by up to 30 percent
(Smartvid.io).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side. Smartvid.io is a vendor of machine-vision-based safety software. Various other
companies (e.g., Autodesk) provide risk assessment solutions (e.g., to manage water risk based
on location and time of year), although in many cases they are not enabled by AI.

Data availability:
The application requires large sets of footage to derive insights on potential safety
hazards. Depending on the level of digitization among contractors and key stakeholders for
developments, these may already be generated and aggregated as part of daily operations. In
this case, the application simply needs to be installed on top of the already available data stack.
In all other cases, video and photo equipment will have to be installed to record footage of the
site.

Technological infrastructure:
Since off-the-shelf-solutions are available that run on any desktop or in the cloud, there are no
specific requirements for technological infrastructure except for cameras, that record images.
The application integrates with common management software suites as well as image and

48
video handling programs (e.g., Procore, Autodesk BIM360, Egnyte).

Capital requirements:
Capital requirements are low and limited to the subscription fee of the software suite used
(plus additional camera equipment if operations are not already documented using camera
equipment). Smartvid.io has an unlimited user pricing model which helps make everyone
responsible for making observations, both positive and risk based (i.e., all workers may
download and use the app without incurring additional cost). However, their pricing plans scale
with the volume of images to be tagged and analyzed (i.e., how many images feed into the
application every month is a cost-determining factor)

Talent requirements:
The application poses no specific requirements for human capital or talent. The software suites
are easy to use and have intuitive interfaces. They do not require any additional training.

Organizational and cultural challenges:

The primary obstacle to overcome in this context is the workers’ willingness to participate in the
reporting and highlighting of safety hazards. While images recorded as part of daily operations
are tagged automatically, workers may also proactively flag safety hazards by uploading them,
which requires a common understanding of the importance of risk mitigation and workplace
safety.

Job impact:
This application has no impact on jobs.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• US-based Suffolk Construction Company implemented Smartvid.io solution over a three-

year period. As a result, 20 percent of all incidents were predicted with an 80 percent
accuracy (using personal protective equipment compliance as a predictor, giving a “level of
compliance”). If only 25 percent of the predicted incidents were avoided, a company with
50 projects a year could prevent 40 to 100 incidents a year. At a cost of approximately EUR
30.000 per incident, this translates to over EUR 1 million in cost savings (Smartvid.io b).
• Shawmut Design and Construction, a Boston-based construction company that already
leveraged construction management and field photo documentation solutions (Procore,
OxBlue), used Smartvid.io’s solution to analyze vast amounts of data and generate

49
 dashboards that helped rank projects by individual risk factors. The company is convinced
that incidents can be prevented from looking at photos alone (Smartvid.io c).

2.19 Automating procurement contract

management by using semantic text analytics
Description:
Semantic text analytics can be used to assist procurement and legal teams with managing
contracts and subcontractor agreements. The applications draw on machine vision and natural
language understanding to extract legally relevant information (e.g., contract length, penalties)
from contracts, agreements, and bids. By semantically analyzing the contents, such applications
can automatically review contracts, warranties, and obligations to compare clauses and bids
in different contracts or to flag clauses which pose specific business risks (e.g., overly high
penalties, non-preferable payment conditions) and therefore need to be reviewed by special
legal counsel. These applications can be particularly relevant for SMEs, as their procurement
functions often have limited legal resources (Sinclair, 2019).

Economic impact:
The economic impact is primarily realized in the form of efficiency gains for procurement and
legal teams, where contract review is greatly sped up. Within legal teams, contract review times
have been observed to drop by over 40 percent. Users of applications like Kira report a 70 to 85
percent accuracy level for contracts automatically processed with the application, freeing up
large amounts of resources for comprehensive quality control (Kira, n.d. a).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do require some degree of customization
from the company’s side. Automated contract analysis is offered for example by AnalyzeLaw
(Germany), Kira (Canada), and eBrevia (US). They cover a wide range of contractual clauses
and are not exclusively focused on the construction industry. However, they do support clauses
specifically applicable to the industry such as supplier clauses, leasing agreements, etc.

Data availability:
All data required to run these applications lies within the contractual documents to be analyzed,
which can be fed into the application in any file format (even scans of third-party forms). No
external or data sources are required.

Technological infrastructure:
Since most of these applications are off-the-shelf solutions that can be run on any computer or

50
in the cloud, no specific technological infrastructure is required.

Capital requirements:
Capital requirements are low and limited to the subscription fee of the software suite used.
Pricing is based on volume of usage and is sold in annual license contracts.

Talent requirements:
These applications do not require enterprises to have specific human capital or talent. The
software suites are easy to use, with intuitive interfaces and a good user experience. They do
not require any substantial additional training.

Organizational and cultural challenges:

The primary obstacle to overcome in this context is acceptance from procurement and legal
departments. They may not fully trust software applications to thoroughly understand all the
nuances of complex legal documents, and since contract management plays an important role
in mitigating business risk, employees may be initially reluctant to rely on these applications in
the context of their daily work. However, where such applications are employed as an additional
source of information for contract review, or as tools to help with the analysis of specific
clauses or to compare contractual offers, they may eliminate laborious manual tasks and
greatly improve employee experience, thus enabling greater adoption and acceptance.

Job impact:
There might be a slightly negative impact on jobs, as procurement departments may handle
their contract work more efficiently, leading to reduced workload and a reduced need for labor.

Environmental sustainability:
No direct or indirect environmental effect is expected.
Examples:

• According to Kira, a solution provider, 4 out of 5 UK magic circle law firms, a third of the
top 100 US law firms, and many US Fortune 100 companies use its software suite. Kira
claims its software has been used in hundreds of billions of dollars of transactions to date
(SeeUnity, n.d.; Kira n.d. b).

51
2.20 Automating lead scoring and prioritizing
customers by analyzing the likelihood to buy
Description:
Businesses may receive hundreds of leads at a time. For sales reps it is hard to differentiate hot
from cold leads without additional sources of insight. Thus, time and energy are wasted on the
wrong leads while viable opportunities may be overlooked. This application aims at assisting
sales reps with prioritizing leads according to their likelihood to buy. The tool leverages AI to
predict probability, based on historical data for various metrics. These may include a lead’s
location, prior engagement with the business, such as website visits, receipt of marketing
emails, or event participation. Measuring the engagement behavior of the customer helps
gauge the buying signals and prioritize leads. The application is particularly relevant for SMEs
as their marketing and sales functions typically can draw on fewer resources than those in their
enterprise counterparts.

Economic impact:
The exact economic impact is a function of customer base, products sold, size of the company,
marketing and sales resources, etc. However, Zoho claims that without a lead management
system, 70 percent of leads fail to convert and that the implementation of such a system leads
to a 47 percent increase in purchases. (Zoho, n.d. a)

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side. Vendors offering intelligent lead scoring and prioritization tools include Zoho,
Freshworks, and Kreato. In all these cases, the application is integrated as a product feature into
the vendor’s CRM software suite.

Data availability:
Many data sources may feed randomly into the application, as nurtured leads are ideally
enriched with as many buying signals as possible. The application may monitor the twitter page
of the company, add users as potential leads if they retweet more than three product posts,
and pass this information on to the sales rep. As the application is typically integrated into
larger CRM systems, CRM data feeds into the prediction. Sales reps do not need to perform any
manual data aggregation tasks.

Technological infrastructure:
As off-the-shelf-solutions exist that can be run on any computer or in the cloud, there are no
specific requirements with regard to technological infrastructure (the application is already

52
integrated in a fully functional software suite; thus, there are no integration risks).

Capital requirements:
Costs are low and limited to the subscription fee of the software suite used. Pricing plans for,
e.g., Freshworks start at EUR 49 per month and user. Ideally, all sales reps in the team would
have their own license, so that monthly costs depend on the size of the company.

Talent requirements:
The application poses no specific requirements to the enterprise with regard to human capital
or talent. The software suites are easy to use and user-friendly, with intuitive interfaces. They do
not require any additional training.

Organizational and cultural challenges:

No significant organizational or cultural challenges are expected as the application serves as an
additional source of insight to sales reps and assists with the initial ranking and prioritization of
leads only. It does not disrupt traditional sales workflows or interfere with the sales reps’ core
task of engaging with customers.

Job impact:
The potential impact on jobs is unclear. There might be a slightly negative job impact, the
magnitude of which depends on the share of the sales reps’ capacity that is freed up and used
to perform high-quality customer engagements.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Finders international, a UK-based family business selling spa and skin care products, used
the CRM solution by Freshworks to achieve an improvement by a factor of two in lead
targeting and by a factor of three in lead qualifying efficiency. (Freshworks, n.d. a)
• Freshworks, a solution provider, claims more than 150,000 customers worldwide use their
CRM solution, including, Bridgestone, TeamViewer, Cisco, Hugo Boss, and numerous SMEs.
(Freshworks, n.d. b)
• Zoho CRM, a solution provider, has over 150,000 customers worldwide, including, Bose,
Netflix, Amazon India, LOGITEK, Ducati, and numerous SMEs. (Zoho, n.d. b)

53
2.21 Automating expense accounting by using visual
recognition
Description:
Expenses are the second largest controllable budget after payroll, yet expense accounting is
still largely performed manually. End-of-month expense reporting poses a significant burden
for all employees involved: those who (manually) enter detailed accounts of their spending
into reporting systems to be reimbursed, accountants who manually review and check them
for eligibility/compliance, and payroll teams who issue remittances for completed expense
reports. The application aims to automate the expense reporting process end to end by allowing
employees to scan relevant receipts, upload them to a shared drive, or simply forward them
to the application as an email. AI helps extract all relevant spending information (including
applicable taxes, exchange rates or beneficiaries) and directly feeds the information into an ERP
system (e.g., accounting, payroll, CRM, HRM). The application could be particularly attractive to
SMEs that don’t have dedicated expense reporting teams and help them comply with local fiscal
authorities without deploying human resources.

Economic impact:
The economic impact comes in the form of valuable time that is given back to employees who
file their expenses at the end of the month. Expense reporting processing times have been
observed to drop by up to 70 percent (Jenji).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side (e.g., Jenji (France), Veryfi (US), and Expensify (US)).

Data availability:
The data required to run the application (i.e., expense data) is supplied by the end user (a firm
employee claiming an expense/reimbursement) through a scan of receipts from which the
application automatically extracts relevant expense management information.

Technological Infrastructure:
Since off-the-shelf-solutions run on any desktop PC or in the cloud, there are no specific
requirements for technological infrastructure. However, if the expense reimbursement process
is automated end to end (i.e., including all steps from digitally extracting the receipt information
from a scan, to checking eligibility of reimbursement and compliance with laws/company
policies, to issuing a remittance), further integration with in-house ERP, payroll, and accounting

54
systems is required. Vendors such as Jenji provide the application programming interfaces
(API) that enable such integrations for many systems, although the exact requirements may
vary by company and legacy system in use.

Capital requirements:
Capital requirements are low and limited to the subscription fee of the software suite used only.
Pricing plans for Jenji, for example, start at EUR 5 per user and month for standard plans. Prices
may differ for large enterprise customers who demand technological integration with their ERP
systems pricing.

Talent requirements:
The application poses no specific requirements for human capital or talent. The software suites
are easy to use and have intuitive interfaces. They do not require any additional training.

Organizational and cultural challenges:

Vendors such as Jenji highlight that their solution has been embraced widely by employees
of their corporate customers who are relieved from the burden of manual expense reporting.
Thus, the application may have a positive impact on the perceived organizational efficiency and
innovation among employees and may help increase organizational health. However, as with all
large-scale system integrations, glitches, exceptions not accounted for, and non-compatibility
with legacy systems may cause frustration among stakeholders.

Job impact:
The introduction of this application could have a negative impact on jobs, as significant
amounts of human labor in expense accounting may be automated.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Example:

• Algolia, a SaaS search specialist company (European HQ in Paris), with 300 employees
and approximately 680 monthly expense receipts, implemented the Jenji solution for
approximately 100 users. The time to process monthly expenses was reduced by 67 percent
in and, as a result, the manual expense management process was completely abandoned.

55
2.22 Improving project performance by predicting
success factors
Description:
Complex projects vary with regard to a wide range of factors (e.g., size, duration). This makes
it difficult to find out why some projects are highly profitable while others underperform.
Many engineering and construction companies rely on leadership judgment that can be
flawed or vulnerable to bias in the context of project management. This application aims at
identifying commonalities among projects with outlier performance (underperforming as
well as overperforming) and determining the factors with the greatest influence. To do so, the
application uses machine learning, advanced analytics, and multivariate analyses to derive
insights and predictors from past projects. Typical factors that are being analyzed include,
e.g., market environment, purchasing, competition, execution, or contractors involved. This
way, a construction company may find that projects of size X, where the number of initial
bidders exceeded Y, yielded low margins in the past. Thus, it may make sense to refrain from
engaging in them in the future. This is particularly relevant for small construction companies
that are more severely hit by losses (i.e., a single project with negative margins) than large
conglomerates that can diversify risk more easily.

Economic impact:
The economic impact of this application is not directly quantifiable and strongly depends on the
insights generated by the analysis and the actions taken by executives in response. Thus, the
impact of individually identified margin drivers, such as “number of bidders for a project,” “joint
venture or solo construction venture,” or “public, private, commercial customer” vary according
to the company, region, size of contract, etc.

Availability of off-the-shelf solutions:

There are currently no off-the-shelf solutions available for this application. This is because
the data sets to be analyzed vary significantly from one construction company to another and
regarding their specific role in the context of a development (e.g., a contractor that engages five
further subcontractors may evaluate project performance in a structurally different way than the
principal developer, who has the end-to-end responsibility). Typically, an in-house data scientist
would carry out the analysis, using tools for statistical modeling and multivariate analyses
(R, SPSS, Python). Design and interpretation of the analysis would require collaboration with
internal stakeholders (to decide on the data sets used, the parameters to be analyzed, etc.).

Data availability:
The data required for running the application is a function of the predictors to be analyzed (e.g.,

56
past procurement contracts may feed into the analysis if it is decided that procurement may
be an influencing factor). However, it tends to be highly fragmented and tedious to aggregate
data in a structured way, which is primarily due to the overall low level of digitization in the
construction industry.

Technological Infrastructure:
There are no specific requirements regarding technological infrastructure. The application can
run on any computer or in the cloud.

Capital requirements:
The costs associated with the application are limited to human capital expenses for a data
scientist with experience in multivariate modeling, and the licensing fees for statistical software
(e.g., SPSS). The data analysis could theoretically also be performed with open programming
environments (e.g., Python, R).

Talent requirements:
The application will likely require the involvement of a data scientist with expertise in
multivariate modelling to become operational. Once set up, the analysis would likely not be
updated with high frequency, thus…

Organizational and cultural challenges:

In an industry exhibiting low levels of digitization, making leaders and executives aware of the
benefits and possibilities of implementing advanced analytics to improve project performance
may pose a great cultural challenge. This is especially due to the fact that running the
application requires significant stakeholder involvement from all levels to discuss hypotheses,
aggregate data, and evaluate the representativeness of data sets.

Job impact:
This application has no impact on jobs.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Multiple research efforts focus on investigating the profitability of construction projects

using big data and machine learning such as (Bilal, November 2019).

57
2.23 Improving HR processes by using HR bots to
automate employee interactions
Description:
Most of the questions an employee could direct to the HR department will typically have
already been answered before in a similar fashion. Specifically, every time a new employee is
onboarded, batches of similar questions (e.g., about benefits, insurance, PTO) will typically
reach the HR department, which is tasked with addressing them in a timely manner to avoid
employee dissatisfaction. This use of AI aims to assist HR professionals by leveraging chatbots
to automate interactions on common HR topics and to deflect inbound inquiries with self-
service options for employees. It draws on natural language processing technology and can be
particularly useful for SMEs, as their HR functions usually have limited resources.

Economic impact:
Economic impact is realized primarily in the form of efficiency gains for HR departments and
increased employee satisfaction. HR departments need to spend less time on repetitive tasks.
According to various studies, the costs associated with manual data entry performed by HR
professionals are approximately EUR 4 per item entry on average (with benefit enrollments
being the most cost-intensive type and onboarding the least expensive). Almost all costs can be
attributed to labor, so significant savings can be expected (Paycom, 2020).

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available (Botcore (UK), Zeal (US), and Servisbot (US)) which do
require some degree of customization from the company’s side.

Data availability:
The data required to run these applications is accessible to HR departments but might need
to be documented and digitized. To set up the chatbot, documents concerning HR matters
(e.g., employee handbooks, benefits programs, company policies) need to be uploaded (in
unstructured format is fine) in order for the chatbot to learn from them. Other vendors offer
prepackaged knowledge bases for HR matters, which can be enhanced by the relevant and
specific company policies.

Technological infrastructure:
Since most of these applications are off-the-shelf solutions that can be run on any desktop PC
or in the cloud, no specific technological infrastructure is required. However, company policies
may require that the application be accessed through the company intranet, in which case an
interface is required.

58
Capital requirements:
Capital requirements are low and limited to the subscription fee of the software suite used.
Monthly billing can range from approximately EUR 3 per month and employee in the case of Zeal
(though this is not for the full range of capabilities of their self-service chatbot), to customized
prices which vary by company size and the range of topics addressable by the chatbot
(sometimes without taking the number of actual users into account).

Talent requirements:
These applications do not require enterprises to have specific human capital or talent. The
software suites are easy to use, with intuitive interfaces and a good user experience. They do
not require any additional training.

Organizational and cultural challenges:

Once the chatbot has been formally introduced internally, there are no significant obstacles
to overcome in this context. Employee acceptance will likely come down to the quality of
interaction with the chatbot and the accuracy of responses. For sensitive or uncommon HR
matters, it is expected that employees will refrain from using the bot and instead engage with
HR directly. If possible, the bot may also directly forward queries to a suitable HR contact and
assign an agent.

Job impact:
There might be a slightly negative impact on jobs, as HR departments may become slightly
more efficient, lowering FTE requirements and overhead.

Environmental sustainability:
No direct or indirect environmental effect is expected.

Examples:

• According to IBM, a technology company, which conducted an HR survey in 2017 of over

8,000 employees, 62 percent of respondents would “reuse a cognitive-enabled tool like a
chatbot to support personal HR decision-making” (IBM, n.d.).

• Botcore, a solution provider, has customers of all sizes and geographies, ranging from
multinational corporations such as Coca-Cola and Unilever to small entities like the West
Midlands Fire Service (Botcore, n.d.).

59
2.24 Optimizing payments by analyzing invoices with
natural language recognition
Description:
Invoice processing and payment is a labor-intensive process since invoices (e.g., from suppliers
or for travel expenses) have to be checked manually line by line. For example, verifying an
expense claim may take an hour or more, especially if there are red flags, such as missing
receipts (Maurer, August 2019). Even when checking every item by itself, duplicates are not
always caught, and issues such as overcharging and nonadherence to agreed payment terms
are difficult to track. AI can help by automatically analyzing invoices and flagging noncompliant
invoices/expense items for humans to verify while automatically paying out the ones that are
correct. Based on the invoice data and contextual information (e.g., contracts, policies), the
AI algorithm uses deep learning techniques to continuously improve its capability of flagging
risky items by learning from human feedback. This includes, for example, identifying duplicate
charges, missing discounts, and suspicious activity, as well as enforcing payment terms,
eliminating overcharges, and flagging compliance issues. Thereby, employees are relieved
from the burden of performing repetitive and monotonous tasks and have more time to focus
on critical issues (e.g., investigating items flagged as risky). Moreover, this is likely to reduce
expenditure (e.g., on items that are noncompliant or double billed) and increases transparency
for managers. As the process is sped up, employees can also be compensated faster. Many
SMEs still process invoices manually on paper; this application could help them to significantly
improve efficiency and remain competitive.

Economic impact:
The human workload for invoice reviewing is reduced by more than half (Maurer, August 2019)
since 100 percent of expenses and invoices can be checked automatically, compared to less
than 10 percent of invoices being checked automatically without an AI solution (AppZen, August
2019). Additionally, expenditure is reduced, as false payments or overpayments are avoided.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do require some degree of customization from
the company’s side (e.g., appzen, oversight systems). Solutions are equipped to flag invoices
(e.g., detecting duplicates or compliance issues). Customization is necessary because these
systems require integration into the specifics of a business (e.g., the policies in place), including
IT system Enterprise Resource Planning (ERP), and must be trained on the contextual data of
the respective business (e.g., supplier contracts).

60
Capital requirements:
Costs largely depend on the data (e.g., invoice or contract data) and systems (e.g., ERP) that
have to be integrated. Costs are moderate if all relevant data is digitized and accessible (e.g.,
Airbus invested around EUR 45,000 (Maurer, August 2019)) and higher if the data has to be
digitized first or made accessible via new tools.

Data availability:
Two types of data are needed for this application – invoice data and contextual data to verify
invoices/assign a risk rating (both internal and external). Ideally, the invoice data is already
digitized, otherwise, visual recognition and natural language processing can be used to read
data off paper and enter it into the system (see Chapter 2.16). To verify invoices and assign
a risk rating to prioritize checks by finance employees, contextual data is needed. Depending
on the setup, the IT systems in place, and specific need (e.g., level of legal compliance), this
may include policies, regulations, contracts, and other internal data such as messaging, badge
access, and system logs. These are internally available and usually already digitized and thus
only need to be entered into the system. Moreover, external data from online sources (e.g.,
on foreign state-owned businesses, politically exposed individuals or entities, and healthcare
professionals) may serve to uncover wasteful, fraudulent, or out-of-policy transactions. Such
data is typically included in the solution.

Technological infrastructure:
This application requires digitized invoices. Therefore, at least one invoice management tool is
necessary and, ideally, visual recognition is in place to automatically read invoices. Furthermore,
invoices need to be put into context with contracts, for example, which must be available via a
digital tool. Notwithstanding these requirements, the application runs on any computer or in the
cloud.

Talent availability:
This application requires integration with existing systems by experts from the provider. Once
operational, the application runs on its own. However, new integrations are required when input
changes (e.g., IT system updates).

Organizational and cultural challenges:

When the AI solution is first introduced, it may be challenging to convince financial staff that the
data processing (nearly) always yields correct results. However, this is likely to change when
staff see that the system reliably points out risks and reduces their workload.

Job impact:
The introduction of this application could have a negative impact on jobs, as the workload of

61
checking invoices and expenses is significantly reduced.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

• Airbus, a European aircraft manufacturer, introduced AI to optimize its expense

reimbursement process. This resulted in significant savings due to labor reduction and
an increase in rejected expense payments: in 2019, the system saved approximately EUR
90,000 and is expected to save about EUR 180,000 in 2020. The system matches reports
against a repository of accepted vendors, expense types, and amounts to spot anomalies.
It examines items such as the identity of a vendor, the existence of venues such as a
restaurant, the type of expense, and the amount. The average time between submission to
approval has decreased from a couple weeks to a few days, and the workload for human
reviewers has been cut by more than half. Over time, the system is expected to detect
patterns to process information faster and more accurately (Maurer, August 2019).

62
2.25 Automating CV screening by using visual
recognition
Description:
Manual screening resumes is a time-consuming part of recruiting. Ideal, a provider of recruiting
automation, estimates that 75 to 88 percent of the CVs received for a role are unqualified. This
leads to high workloads associated with going through resumes and shortlisting candidates,
consuming up to 23 hours of a recruiter’s time for a single hire. (Ideal, n.d. a)
This AI application aims at significantly increasing the speed at which inbound job applications
can be processed. For this, it uses two sources of input. First, the applicant target profile
is defined by recruiters and the specific requirements of the job posting. Second, all job
applications are fed into the tool in a machine-readable format (e.g., pdf, word; if a machine-
readable format is not available, a preceding step of scanning and optical character recognition
could be introduced). The tool subsequently uses natural language understanding and semantic
analysis (among other AI technologies) to extract relevant information from resumes and
other supporting documentation (e.g., work experience, educational qualifications, skills and
knowledge, competencies, personality traits). The extracted information is then compared with
the requirements from the job application, and the best suited applicants are suggested to the
recruiter. The application is particularly relevant for SMEs as their HR functions can usually draw
on fewer resources than those in their enterprise counterparts.

Economic impact:
The economic impact is manifold. Automating the CV screening process frees up recruiters’
capacity and thus reduces screening cost directly (by up to 75 percent, according to Ideal).
Furthermore, recruiters’ capacity can now be shifted to higher-value-adding tasks, such as
getting to know candidates better. Building relationships and determining the cultural fit
may ultimately lead to a higher quality of hiring, which is likely to have an effect on corporate
success (though the associated economic gains are harder to quantify). Additionally, the tool
helps eliminate unconscious bias throughout the hiring process as, e.g., recruiters might favor
applicants from one university because they know this particular institution very well.

Availability of off-the-shelf solutions:

There are off-the-shelf solutions available which do not require further customization from the
company’s side (e.g., Ideal (Canada) and CiiVSOFT (UK)).

Data availability:
There are two types of data needed for this AI application: the job applications from the
candidates and the job posting of the company. All this lies within the applicant tracking

63
systems and recruiting stacks already in place (e.g., Workday, SuccessFactors, Greenhouse,
Bullhorn), and is thus readily accessible. No external data sources are required; however,
recruiters need to input the type of candidate profiles they deem relevant for the position into
the application up front, so that the software can apply relevant filtering and accurate ranking.

Technological infrastructure:
As off-the-shelf-solutions exist that can be run on any computer or in the cloud, there are no
specific requirements with regard to technological infrastructure. Most off-the-shelf solutions
seamlessly integrate with common recruiting stacks and thus do not disrupt workflows.

Capital requirements:
Expenses are low, limited to the subscription fee of the software suite used, which increases
with the size of operations. Pricing plans for, e.g., CiiVSOFT start at approximately EUR 500/
month for the standard plan, which allows for 1,000 applicant rankings per month. This may be
sufficient for the typical SME.

Talent requirements:
The application poses no specific requirements to the enterprise with regard to human capital
or talent. The software suites are easy to use and user-friendly, with intuitive interfaces. They do
not require substantial additional training, and the tools integrate well with standard applicant
tracking systems.

Organizational and cultural challenges:

No significant organizational or cultural challenges are expected as the application serves as
an additional source of insight to recruiters, and assists with the initial ranking of applicants
(which recruiters can override throughout the workflow, should their perspective differ). As the
application may help find candidates who exhibit a particularly strong cultural fit, organizational
health may improve overall.

Job impact:
The introduction of this application could have a slightly negative impact on jobs. The
magnitude of which depends on the share of the recruiter’s capacity that can be shifted from
repetitive to higher value-adding tasks.

Environmental sustainability:
No direct or indirect environmental effects are expected.

Examples:

64
• Indigo, an American book and music retailer, used Ideal to support their weekly influx
of 2,200 online applications. By using AI, Ideal helped Indigo reduce cost per hire by 71
percent, triple the qualified candidates, remove potential for hiring bias, and improve
recruiter efficiency by a factor of 3.7. Candidate waiting times dropped significantly, and top
candidates were quickly identified and contacted. (Ideal, n.d. b).

65
3 Appendix
3.1 Cloud computing offers SMEs the possibility
to dynamically and economically scale
computation, storage, and network capacities

Cloud computing is a new computing paradigm in which dynamically scalable resources

(e.g., computing, storage and network) are provided as a service on a pay-per-use basis. Key
characteristics and advantages over traditional server infrastructure are (not exhaustive):
cost to enterprises.

Resource pooling
Through virtualization/hardware abstraction customers can share resources and thus achieve
economies of scale unseen for legacy IT environments (e.g., it is cheaper for Amazon Web
Services (AWS) to offer computational capacity to its thousands of customers at the same time
than it is for a Fortune500 company to set up their own data center).
• Automation/rapid elasticity
Through hardware abstraction cloud customers can easily scale up or down the
computational resources they require (e.g., Netflix as a cloud customer may need more
resources during peak hours where most people watch, but can scale down its servers
during less busy hours, these servers can then be used by someone else (see pooling
above)).
• On-demand self-service
Through self-provisioning customers can dynamically scale the computational resources
they need, even in cases where demand spikes cannot be foreseen. E.g., if an ecommerce
company hosts their web shop on-cloud on Black Friday, the company could be allocated
more CPUs/servers instantaneously and automatically to prevent the server from breaking
down.
• Metered/measured service
In stark contrast to legacy IT environments, cloud costs come as opex- or usage-based
billings. This leads to significant cost savings, as on-premise data centers which have
to be operated, maintained and provisioned around the clock regardless of actual usage
constitute high fixed cost to enterprises.

66
3.2 The General Data Protection Regulation (GDPR)
protects personal data and allows the free flow
of data within the EU6

The General Data Protection Regulation (GDPR) came into effect in May 2018. It protects
the personal data of EU citizens and residents, defining and establishing fundamental
data protection principles that concern the full process of collection, processing, use and
responsibility around personal data.7 In addition it formalizes and allows free flow of data within
the EU, which is an important prerequisite for the adoption of artificial intelligence (AI). GDPR
is considered to be the strictest data privacy and security law in the world, according to the EU
(GDPR.EU, n.d.). The violation fines are set at a maximum of EUR 20 million, or 4 percent of the
global revenue of a company if it is higher, excluding potential compensations to data subjects
(person whose data is processed). GDPR has implications for the entire supply chain of data
processing, from data gathering, to storage and use.

The regulation applies to personal data, which are defined as any information that relates to
an individual who can be directly or indirectly (GDPR.EU, n.d.). Typical examples of personal
data are e-mail address, mobile phone number, social security number or name, surname, birth
date and address. However, the GDPR regulations do not apply to completely anonymized data
or aggregated data where linking data to individuals is impossible (this does not hold true for
pseudonymization where linking data to individuals is still possible, e.g., through the use of a
(secret) identifier even if this identifier is only known to a select group of people).

The applicability of GDPR regulations covers different types of entities such as companies,
institutions, organizations, universities, etc. Furthermore, the applicability is not limited to
entities located in the EU. Any entity processing data of EU residents, or active in supplying
goods and services to EU consumers, needs to comply with GDPR, even if the data processing
happens outside of the EU. It is important to note that some exceptions from the otherwise
strict GDPR regulations exist. For example, the use of personal data for national and public
interest (e.g., policing activities, prosecution, etc.) is not as strictly regulated (e.g., no consent
for data gathering or processing is needed).

6 This document does not render any legal advice. The descriptions given in this document do not constitute nor substitute
seeking legal advice.
7 Among the key principles are data minimization (data gathering, storage and use should be limited to the extent necessary for
the purpose), storage limitation (store personal data only for as long as necessary), accountability (data controller’s responsibility for being
able to demonstrate compliance), purpose limitation (data processed for the purposes specified to data subject when collected), integrity
and confidentiality (e.g., data processing must ensure appropriate security, e.g., through encrypting), lawfulness (data processed only if
there is a lawful basis for it, e.g., consent), fairness (data subjects should have means to exercise their rights) and transparency (information
for data subject should be easy to understand), and accuracy of data (possibility of timely rectification of wrong data).

67
The regulation defines key legal terms for further reference. Data subjects are individuals whose
data is processed (e.g., the consumer). The data controller is the organization (e.g., the SME)
who decides about “why” and “how” personal data will be processed. At last, a data processor is
the entity that processes the data on behalf of the data controller, which can be done internally
or externally, via 3rd parties on a contractual basis.

Under GDPR, there are six lawful grounds for data gathering, storage and usage:
1. Legal obligations (e.g., judiciary system, taxes)
2. Vital interest (e.g., physical integrity of the data subject)
3. Contractual obligations (e.g., processing the address of a data subject during online
purchasing)
4. Public task (e.g., organization exercising official authority to carry out a task in the public
interest)
5. Legitimate business interest (e.g., direct marketing to existing customers)
6. Consent (e.g., the data subject giving consent to data gathering and/or storage and/or use)

For business the latter two will constitute the lawful ground for the majority of cases of data
gathering. While the legitimate business interest does not require consent, it can only use the
data of existing customers and this data cannot be transferred to 3rd parties. Alternatively,
businesses can ask the data subject for consent. This consent has to be informed (i.e. written in
understandable language and not as part of overarching terms and conditions) and affirmative.
Consent for data sharing with external partner must be collected if sharing of the data is
needed. Furthermore, an additional personal consent must be collected for sensitive data in
relation to fundamental rights and freedoms (e.g., health data, religion, political opinion, and
race). This data should not be used for business if it can be avoided, as additional security
measures are needed (e.g., a DPIA, see below).

During the data usage phase, there should be a record of processing activity, inclusive initial
purpose, time limits of usage and security safeguards. The data should be processed in a
transparent manner, meaning consent was provided, activities are recorded, and the company
must be able to inform both legislators and data subject of how data is handled, if this is
requested by either one. After data usage, the data processor has to delete and return all
personal data to the data controller. The EU prohibits transmission of data to non-EU countries
without consent as well, unless adequate level of protection is ensured by these countries. As
of February 2020, a few European countries fulfill minimal level of personal data protection,
among them Norway, Lichtenstein and Iceland, and Switzerland. The United States’ compliance
is limited to the Privacy Shield Framework.8

8 The Privacy Shield Framework protects the fundamental rights of anyone in the EU whose personal data is transferred to the
United States for commercial purposes, with safeguards on US government access. The free transfer of personal data can only

68
In terms of storage of personal data, the data should be kept in a form that enables
identification of individuals for no longer than necessary. In case there is a high risk to people’s
personal information, a Data Protection Impact Assessment (DPIA) procedure is needed for
each of the data processing activities, including highly sensitive data. Breaches must be notified
to the authorities and data subjects within 72 hours if it cannot be shown that the appropriate
technological measures were taken (e.g., encryption). Concretely, this means that companies
will have to develop a “Data Breach Reporting and Action plan”. Under certain conditions9, the
company has to name a Data Protection Officer (DPO), whose role will be to inform and advise
other employees, monitor compliance with the regulation, provide advice about the DPIA,
cooperate with supervisory authority and act as point of contact with them. Data protection
could be achieved through e.g., two-factor authentication to access to accounts where data are
stored to work with it, limited access to employees on a need-to-know-basis, and end-to-end
encryption with cloud providers. Privacy should be embedded upfront in processes (“Privacy-by-
design”).10

Data subjects have extensive rights, notably the right to have access to the data processed, be
informed about how the decision algorithms works, obtain rectification or erasure of personal
data under certain circumstances (“right to be forgotten”), and object or restrict data processing
(e.g., revoke consent).

As many AI applications require the gathering, storage and use of personal data (see below),
SMEs should take the following steps to ensure compliance with GDPR:
• Have at least one qualified person following a training, or get advisory services from one of
the numerous suppliers on the market (iapp, 2019).
• Establish whether data used in the AI application is covered by GDPR (i.e. whether it is
considered personal data), and what risks are associated with processing these. For
example, the use of sensitive data (as described above) should be avoided, if it is not
absolutely needed for the business purpose. Understand whether the types of data are
within GDPR regulation, as many types of data such as machine data, process data, storage
data, etc. can be used without complying to GDPR regulations.
• Ensure technical protection (e.g., encryption) of the data and do not share data without the
necessary precautions. If shared, data processors need to be GDPR compliant.
• Be aware of obligations towards regulator and customers (e.g., consent, information
procedure for customer including data deletion possibilities).
• Be able to explain the function of AI algorithms and their decision to both customer and
regulator how the decision is being made.

happen if companies are certified in the US under the Privacy Shield.

9 A DPO is needed in when the processing of data is carried out by a public authority or when core activities of the controller and
processor require regular and systematic monitoring of data subjects on a large scale.
10 End-to-end encryption and privacy should be the default setting, with transparency in processes and technology.

69
• In addition to GDPR, more AI-specific regulation is expected, as this is one of the priorities of
the newly elected European Commission, with focus on ethical use of AI (Baker 2019).

4 Sources
- 365FarmNet. (n.d.) 365FarmNet solutions support you in the management of your farm.
Available from: https://www.365farmnet.com/en/prices/ [Accessed 19th February 2020].
- 904labs (n.d.). A.I. for search. Available from: https://www.904labs.com/en/ [Accessed 13th
February 2020].
- ABI Research. (May 2018) A Mixed Reception for Augmented Reality in Bricks & Mortar
Retail: Efficiencies over Engagement. Available from: https://www.prnewswire.com/news-
releases/a-mixed-reception-for-augmented-reality-in-bricks--mortar-retail-efficiencies-over-
engagement-300651752.html [Accessed 10th February, 2020].
- Abousaeidi, M., Fauzi, R., and Muhamad, R. (September 2016). Geographic Information
System (GIS) modeling approach to determine the fastest delivery routes. Saudi Journal of
Biological Sciences, 23(5).
- Abundant robotics. (n.d.) Welcome to the world’s first commercial robotic apple harvest.
Available from: https://www.abundantrobotics.com/ [Accessed 19th February 2020].
- Accenture Interactive. (2016) Chatbots in Customer Service. Available from: https://www.
accenture.com/t00010101T000000__w__/br-pt/_acnmedia/PDF-45/Accenture-Chatbots-
Customer-Service.pdf [Accessed 26th February 2020].
- Afiniti. (n.d.) Clients. Available from: https://www.afiniti.com/clients [Accessed 5th February
2020].
- Aglytix. (n.d.) Homepage. Available from: https://www.aglytix.com/ [Accessed 19th
February 2020].
- Agricon. (n.d.) Digital base fertilization with Agricon. Available from: https://www.agricon.
de/en/base-fertilization/ [Accessed 3rd February 2020].
- Agritec. (n.d.) Advanced Software for Farm Management. Available from: https://www.
agritecsoft.com/en/ [Accessed 19th February 2020].
- AgSquared. (n.d.) Solutions. Available from https://www.agsquared.com/solutions
[Accessed 19th February 2020].
- Aiosell. (n.d. a) Case studies. Available from: https://aiosell.com/case-studies/ [Accessed
5th February 2020].

- Aiosell. (n.d. b) Paya Beach Spa & Dive Resort – Malaysia. Available from: https://aiosell.
com/case-studies/paya-beach-spa-dive-resort-malaysia/ [Accessed 5th February 2020].

70
- Aktar W. et al. (March 2009). Impact of pesticides use in agriculture: their benefits and
hazards. Interdisciplinary toxicology, 2(1), 1–12. doi:10.2478/v10102-009-0001-7.
- Al Ohali, Y. (January 2011) Computer vision-based date fruit grading system – design and
implementation. Journal of King Saud University-Computer and Information Sciences, 23(1),
29–36, https://doi.org/10.1016/j.jksuci.2010.03.003.
- AMAZONE. (n.d.) ZA-M, ZA-M Profis. Available from: https://www.technik-nordwest.
de/media/technik_nordwest_de/partner_in_der_landtechnik/amazone_2/prospekte_1/
Prospekt_ZA-M.pdf [Accessed 3rd February 2020].
- Amplero. (n.d.) BECU Leverages Amplero to Increase Member Engagement, Promote
Financial Health and Achieve Business KPI’s. Available from: https://amplero.com/becu-
and-amplero/ [Accessed 12th February 2020].
- AppZen. (August 2019). The State of AI in Business Spend. Available from: https://
cdn2.hubspot.net/hubfs/516015/White%20papers/The%20State%20of%20AI%20in%20
Business%20Spend%20Report%20Q3%202019.pdf [Accessed 12th February 2020].
- Ashford, W. (March 2019). Small businesses hit hardest by cyber crime costs. Computer
Weekly. Available from: https://www.computerweekly.com/news/252460134/Small-
businesses-hit-hardest-by-cyber-crime-costs [Accessed 12th February 2020].
- Autodesk. (2018) How GM is using generative design for vehicles of the future. Available
from: http://fom.autodesk.com/customer-innovation-spotlight/p/2 [Accessed 12th February
2020].
- Autor D. et al. (May 2017) The fall of the labor share and the rise of superstar firms. IZA
discussion paper nr 10756. Available from: https://economics.mit.edu/files/12979
- Autovista Group (February 2019). The state of autonomous legislation in Europe. Available
from: https://autovistagroup.com/news-and-insights/state-autonomous-legislation-europe
[Accessed 19th February 2020].
- Baker A. (October 2019) Von der Leyen, new Commission take aim at AI legislation. Atlantic
Council. Available from: https://www.atlanticcouncil.org/commentary/blog-post/von-der-
leyen-new-commission-take-aim-at-ai-legislation/ [Accessed 6th February 2020].
- Banker, S. (March 2019) The Autonomous Mobile Robot Market Is Taking Off Like A Rocket
Ship. Forbes.com. Available from: https://www.forbes.com/sites/stevebanker/2019/03/11/
the-autonomous-mobile-robot-market-is-taking-off-like-a-rocket-ship/#1496c8111603
[Accessed 12th February 2020].
- Barracuda (n.d.). Dutch DIY giant shelves email attacks. Available from: https://assets.
barracuda.com/assets/docs/dms/Barracuda_TTP_CS_Intergamma_UK.pdf [Accessed 12th
February 2020].
- BayWa AG. (February 2019) Talking fields. 75|600 The BayWa AG Magazine. Available from:
https://www.7560-digital.com/en/agricultural-trends/talking-fields [Accessed 3rd February
2020].
- Behrendt K. et al. (April 2019) The impact of swarm robotics on arable farm size and

71
 structure in the UK. Agri-Tech Economics Papers. Available from: https://www.researchgate.
net/publication/332653186_The_Impact_of_Swarm_Robotics_on_Arable_Farm_Size_and_
Structure_in_the_UK [Accessed 19th February 2020].
- Belamy, E. (June, 2019) UK small businesses remain vulnerable to cyber-attack. Available
from: https://www.swcomms.co.uk/blog/article/uk-small-businesses-remain-vulnerable-to-
cyber-attack/ [Accessed 12th February 2020].
- Bellodi E. et al. (December 2017). Economic and organizational impact of a clinical decision
support system on laboratory test ordering. BMC Medical Informatics and Decision Making.
Available from: https://bmcmedinformdecismak.biomedcentral.com/articles/10.1186/
s12911-017-0574-6 [Accessed 18th February 2020].
- Bespoke Inc. (August 2017). Hotel Chatbot: Bebot Introduced Across an Emerging Hotel
Brand in Japan. PR Newswire. Available from: https://www.prnewswire.com/news-releases/
hotel-chatbot-bebot-introduced-across-an-emerging-hotel-brand-in-japan-300497618.html
[Accessed 5th February 2020].
- Bilal et al. (November 2019) Investigating profitability performance of construction projects
using big data: A project analytics approach, Journal of Building Engineering, Volume 26.
- Bitrefine Group. (n.d.) Automatic fruits/ flowers sorting. Available from: https://bitrefine.
group/industries/precision-agriculture/88-industries/agriculture-food/agriculture-
solutions/178-automatic-fruits-flowers-sorting [Accessed 19th February 2020].
- Blue River Technology. (n.d.) Blue River Technology – Optimize every plant. Available from:
http://www.bluerivertechnology.com/ [Accessed 4th February 2020].
- BlueYonder (n.d.) Replenishment and Price Optimization at OTTO. Available from: https://
jda.com/-/media/jda/knowledge-center/case-studies/by_case_study_otto_en.ashx
[Accessed 5th February 2020].
- BMW (June 2019). BMW of North America Introduces Smart Glasses. Available from:
https://www.press.bmwgroup.com/usa/article/detail/T0297030EN_US/bmw-of-north-
america-introduces-smart-glasses-and-new-technician-communication-systems-at-u-s-
dealerships-to-reduce-repair-times-and-improve-customer-satisfaction?language=en_US
[Accessed 12th February 2020].
- Bosch. (n.d.) Smart spraying – precision herbicide application on weeds. Available from:
https://www.bosch.com/research/know-how/success-stories/smart-spraying-precision-
herbicide-application-on-weeds/ [Accessed 4th February 2020].
- Boston Globe. (October 2019) Available from: https://www.bostonglobe.com/
business/2019/10/10/prevent-workplace-accidents-construction-companies-are-turning-
vinnie/ZEpSKKj1h2EEYHU0PtMh3H/story.html [Accessed 12th February 2020].
- Botcore. (n.d.) HR Bots. Available from: https://botcore.ai/hr-bots/ [Accessed 18th February
2020].
- Boudet J. et al. (June 2019) The future of personalization – and how to get ready for it.
McKinsey & Company. Available from: https://www.mckinsey.com/business-functions/

72
 marketing-and-sales/our-insights/the-future-of-personalization-and-how-to-get-ready-for-it
[Accessed 5th February 2020].
- Brown et al. (March 2014) Continuous Monitoring in an Inpatient Medical-Surgical Unit: A
Controlled Clinical Trial. The American Journal of Medicine, Volume 127, Issue 3, 226-232.
- Brown et al. (December 2010) Improved Outcomes and Reduced Costs with Contact-
Free Continuous Patient Monitoring on a Medical-Surgical Hospital Unit. Available from:
https://strumedical.com/admin/allegati/464-White_Paper_Patient-safety_Dec8_2010.pdf
[Accessed 26th February 2020].
- C3 Solutions. (n.d.) Dock Scheduling Case Study. Available from: https://info.c3solutions.
com/dock-scheduling-case-study [Accessed 5th February 2020].
- Capgemini. (July 2017) Reinventing Cybersecurity with Artificial Intelligence. Available
from: https://www.capgemini.com/wp-content/uploads/2019/07/AI-in-Cybersecurity_
Report_20190711_V06.pdf [Accessed 6th February 2020].
- Capterra. (n.d.) Optimove. Available from: https://www.capterra.com/p/131771/Optimove/
[Accessed 12th February 2020].
- Cembali, T., Folwell, R. J., Clary, C. D., & Mari, M. (October 2008). “Economic comparison of
selective and nonselective mechanical harvesting of asparagus.” International Journal of
Vegetable Science, 14(1), 4–22.
- Cembali, T., Folwell, R. J., Huffaker, R. G., McCluskey, J. J., & Wandschneider, P. R. (June
2005). “Economic evaluation of selective mechanical harvesting for asparagus.” XI
International Asparagus Symposium 776, 33–44.
- Chen, W. et al. (January 2017) A new method of large-scale short-term forecasting of
agricultural commodity prices: illustrated by the case of agricultural markets in Beijing.
Journal of Big Data. No 4. Available from: https://link.springer.com/article/10.1186/s40537-
016-0062-3#citeas [Accessed 28th February 2020].Choudhury, S. R. (December 2019).
Automated hacking, deepfakes are going to be major cybersecurity threats in 2020. CNBC.
Available from: https://www.cnbc.com/2019/12/18/automated-hacking-deepfakes-top-
cybersecurity-threats-in-2020.html [Accessed 12th February 2020].
- CiTF Winetech Divisio.n (n.d.) Homepage. Available from: https://usa.citf-group.com/en/
solutions-for-wineries [Accessed 19th February 2020]

- Cliffordchance.com. (November 2017). Artificial intelligence and the future for legal
services. Available from: https://www.cliffordchance.com/content/dam/cliffordchance/
briefings/2017/11/artificial-intelligence-and-the-future-for-legal-services.pdf [Accessed
13th February 2020].
- Cobots.ie. (May 2019). Carl Zeiss Case Study: Collaborative Robots Increase Productivity
by 90%. Available from: https://cobots.ie/news/carl-zeiss-case-study-collaborative-robots-
increase-productivity-by-90 [Accessed 12th February 2020].
- Colliers International. (September 2018) The Hotels of the Future: Redefining
Experience & Profits. Available from: https://arabiantravelmarket.wtm.com/

73
 RXUK/RXUK_ArabianTravelMarket/2019/PDF/Colliers/Colliers%20-%20ATM%20
Knowledge%20Partner%20-%20Disruptive%20Technology%20-%20%20English.
pdf?v=636916080286833560
- Comptera. (n.d. a) Price optimization at find me a gift. Available from: https://competera.
net/pdf/CS_gift_en.pdf [Accessed 5th February 2020].
- Comptera. (n.d. b) How a Leading European Retailer Maximized Revenue Without Losing
Margin. Available from: https://competera.net/pdf/case-study-po.pdf [Accessed 5th
February 2020].
- Continental. (n.d.) Back to business: Thanks to the New Agricultural Robot of Continental
Farmers Can Refocus on the Essentials. Available from: https://www.continental.com/en/
press/fairs-events/agritechnica-2019/agricultural-robot-contadino-197232 [Accessed 19th
February 2020].
- Crane-Droesch, A. (October 2018) Machine learning methods for crop yield prediction and
climate change impact assessment in agriculture. Environmental Research Letters. Vol 12.
Num 11. Available from: https://iopscience.iop.org/article/10.1088/1748-9326/aae159/pdf
[Accessed 28th February 2020].
- Croptracker. (n.d.) Croptracker Farm Management Software. Available from: https://www.
croptracker.com/product/farm-management-software.html [Accessed 19th February 2020].
- CrowdStrike. (n.d.) Falcon Endpoint Protection. Available from: https://www.crowdstrike.
com/endpoint-security-products/ [Accessed 6th February 2020].
- Current Health. (n.d. a) Homepage. Available from: https://currenthealth.com/ [Accessed
12th February 2020].
- Current Health. (n.d. b) How Dartford and Gravesham NHS Home Health Used Current to
Reduce Therapy Visits by 22 Percent. Available from: http://go.currenthealth.com/content/
dartford-nhs-case-study [Accessed 12th February 2020].
- Cyberreason. (n.d.) How Cybereason Edged the Competition and Automated Threat Hunting
for a Global Aerospace Manufacturer. Available from: http://go.cybereason.com/rs/996-
YZT-709/images/Cybereason-Case-Study-Global-Aerospace-Manufacturer.pdf [Accessed
7th February 2020].
- Dahikar, S. S., and Rode, S. V. (2014) Agricultural crop yield prediction using artificial neural
network approach. International Journal of Innovative Research in Electrical, Electronics,
Instrumentation and Control Engineering, 2(1), 683-686.
- Davis. (n.d.) Vantage Pro2 - Versatile And Rugged For Serious Weather Watching. Available
from: https://www.davisinstruments.com/solution/vantage-pro2/ [Accessed 4th February
2020].
- Deere, J. (October 2017) Selbstfahrer R4040i / R4050i. Available from: https://www.robert-
aebi-landtechnik.ch/domains/matra_ch/data/free_docs/Serie_R4040i_und_R4050i.pdf
[Accessed 3rd February 2020].
- Deloitte. (July 2018) European Construction Monitor. Available from: https://www2.

74
 deloitte.com/content/dam/Deloitte/pl/Documents/Reports/pl_European_Construction_
Monitor_2017-2018.PDF [Accessed 19th February 2020].
- Demand-Planning.com. (n.d,) How much does forecasting software cost, & how much
will it save? Available from: https://demand-planning.com/2018/07/12/how-much-does-
forecasting-software-cost/, [Accessed 5th February 2020].

- Rayome, A.D. (February 2018) How Sephora is leveraging AR and AI to transform retail
and help customers buy cosmetics. Available from: https://www.techrepublic.com/article/
how-sephora-is-leveraging-ar-and-ai-to-transform-retail-and-help-customers-buy-cosmetics/
[Accessed 7th February 2020].
- Design World. (February 2019). AI at the Edge: How “Cobots” Can Process Sensor Data
Quickly. Available from: https://www.designworldonline.com/ai-at-the-edge-how-cobots-
can-process-sensor-data-quickly/ [Accessed 7th February 2020].
- Dicardology. (April 2019) Medivis Unveils AnatomyX Augmented Reality Education
Platform. Available from: https://www.dicardiology.com/product/medivis-unveils-anatomyx-
augmented-reality-education-platform [Accessed 19th February 2020].
- Dimitropoulos S. (August 2019) If one drone isn’t enough, try a drone swarm. BBC News.
Available from: https://www.bbc.com/news/business-49177704 [Accessed 19th February
2020].
- DJI. (n.d. b) DJI agriculture. Available from: https://www.dji.com/products/
agriculture?site=brandsite&from=nav [Accessed 4th February 2020].
- DJI. (n.d.) AGRAS MG-1. Available from: https://www.dji.com/mg-1 [Accessed 3rd February
2020]
- Doblas-Reyes F. et al. (2003). Weather and climate forecasts for agriculture. Guide to
agricultural, meteorological practices, 57. Chapter 5.
- Doc.arcgis.com. (n.d.). System requirements – ArcGIS Earth. Available from: https://doc.
arcgis.com/en/arcgis-earth/get-started/system-requirement.htm [Accesses 4th February
2020].
- Dorhour R&D. (n.d.) Prospero: The robot farmer. Available from: http://dorhoutrd.com/
prospero_robot_farmer [Accessed 19th February 2020].
- Dronepart.de. (n.d.) DJI Phantom 4 RTK | Drohne für Vermessung. Available from: https://
droneparts.de/dji-phantom-4-rtk-drohne-fuer-vermessung?spartner=8921815&gclid=eaiaiqo
bchmivkytuo-35wivk-f3ch10twnbeaqyasabegibifd_bwe [Accessed 4th February 2020].
- Duetto. (n.d. a) Automation Drives Business Success For Ruby Hotels. Available from:
https://www.duettocloud.com/hubfs/Ruby%20Hotels%20Case%20Study.pdf [Accessed 5th
February 2020].
- Duetto. (n.d. b) Advanced Pricing Central to NH Hotels Group’s Turnaround. Available from:
https://www.duettocloud.com/hubfs/assets/case-study-NH-Hotel.pdf [Accessed 5th
February 2020].
- EarlySense. (n.d. a) Clinical outcomes from earlysense contact-free continuous patient

75
 monitoring. Available from: https://www.earlysense.com/clinical-outcomes/ [Accessed 5th
February 2020].
- Ecologic Institute. (February 2017) Updated Inventory and Assessment of Soil Protection
Policy Instruments in EU Member States. Available from: https://ec.europa.eu/environment/
soil/pdf/Soil_inventory_report.pdf [Accessed 3rd February 2020].
- ecoRobotix Ltd. (n.d.) Technology for environment. Available from: https://www.ecorobotix.
com/en/avo-autonomous-robot-weeder/ [Accessed 4th February 2020].
- Ehnts A. (February 2019) Mit Sensoren Sparsam Spritzen. In Agrarheute 03/2017. Available
from: https://www.agricon.de/blog/2019/02/22/mit-sensoren-sparsam-spritzen/ [Accessed
3rd February 2020].

- Eisenberg, A. (n.d.) How Much Does It Cost to Develop an AR App? Available from: https://
appreal-vr.com/blog/augmented-reality-app-development-costs/ [Accessed 10th February
2020].
- Engineering.com. (November 2018) Generative Design Produces Many Shapes, but
Can You Make Them? Available from: https://www.engineering.com/DesignSoftware/
DesignSoftwareArticles/ArticleID/18082/Generative-Design-Produces-Many-Shapes-but-
Can-You-Make-Them.aspx [Accessed 7th February 2020].
- Epson (n.d.). MOVERIO BT-350. Available from: https://www.epson.de/en/products/see-
through-mobile-viewer/moverio-bt-350 [Accessed 7th February 2020].
- Esri.com (n.d.). GIS in Transportation – Location Intelligence Strengthens Infrastructure Life
Cycle Management. Available from: https://www.esri.com/en-us/industries/transportation/
overview [Accessed 4th February 2020].
- European Commission. (December 2018) Coordinated plan on Artificial Intelligence.
Available from: https://ec.europa.eu/digital-single-market/en/news/coordinated-plan-
artificial-intelligence [Accessed 30th January 2020].

- European Commission. (February 2019). Exchange of electronic health records across the
EU. Available from: https://ec.europa.eu/digital-single-market/en/exchange-electronic-
health-records-across-eu [Accessed 20th February 2020].
- European Commission. (April 2019) Building trust in human-centric artificial intelligence.
Available from: https://ec.europa.eu/digital-single-market/en/news/communication-
building-trust-human-centric-artificial-intelligence [Accessed 30th January 2020].

- European Commission. (June 2019) Digital Europe Programme Factsheet. Available from:
https://ec.europa.eu/digital-single-market/en/news/digital-europe-programme-proposed-
eu92-billion-funding-2021-2027 [Accessed 30th January 2020].
-
- European Commission. (June 2019b) Agricultural Factsheet. Available from: https://
ec.europa.eu/info/food-farming-fisheries/farming/facts-and-figures/markets/production/

76
 production-country/statistical-factsheets_en [Accessed 28th February 2020].
-
- European Commission. (February 2020) White Paper on Artificial Intelligence: a European
approach to excellence and trust. Available from: https://ec.europa.eu/info/publications/
white-paper-artificial-intelligence-european-approach-excellence-and-trust_en [Accessed
25th February 2020].
-
- European Commission. (n.d.) Breakdown of the Horizon 2020 Budget. Available from: http://
ec.europa.eu/research/horizon2020/pdf/press/horizon_2020_budget_constant_2011.pdf
[Accessed 30th January 2020].

- European Parliament (March 2006), REGULATION (EC) No 561/2006 OF THE EUROPEAN

PARLIAMENT AND OF THE COUNCIL of 15 March 2006 on the harmonisation
of certain social legislation relating to road transport and amending Council
Regulations (EEC) No 3821/85 and (EC) No 2135/98 and repealing Council Regulation
(EEC) No 3820/85. Available from: https://eur-lex.europa.eu/legal-content/EN/
TXT/?uri=CELEX:02006R0561-20150302 [Accessed 5th February 2020].
- Eurostat. (2013) European System of Accounts 2010. Available from: https://ec.europa.eu/
eurostat/documents/3859598/5925693/KS-02-13-269-EN.PDF/44cd9d01-bc64-40e5-bd40-
d17df0c69334 [Accessed 26th February].

- Eurostat. (2016) Structural business statistics. Available from: https://ec.europa.eu/

eurostat/web/structural-business-statistics [Accessed 26th February].

- Eurostat. (January 2020a) SME definition Available from: https://ec.europa.eu/eurostat/

web/structural-business-statistics/structural-business-statistics/sme [Accessed 26th
February].

- Eurostat. (January 2020b) National accounts aggregates by industry. Available from:

https://ec.europa.eu/eurostat/web/products-datasets/product?code=nama_10_a64
[Accessed 26th February].
-
- Expertinsights.com (n.d.). Review Ironscales. Available from: https://www.expertinsights.
com/reviews/ironscales [Accessed 7th February 2020].
- Feldschwarm. (n.d.) Feldschwarm® – autonome Feldmodule für den
ressourcenschonenden Landbau. Available from: http://www.feldschwarm.de/ [Accessed
19th February 2020].
- Fendt. (September 2017) Fendt puts the new robot ‘Xaver’ to use. Available from: https://
www.fendt.com/int/fendt-xaver [Accessed 19th February 2020].

77
- Fetch Robotics. (n.d.). Is Your Warehouse Ready for AMRs? The Answer is “Yes.”. Available
from: https://fetchrobotics.com/fetch-robotics-blog/is-your-warehouse-ready-for-amrs/
[Accessed 7th February 2020].
- Fetch Robotics. (n.d.). Ryder Customer Success Story. Available from: https://fetchrobotics.
com/ryder-customer-success-story/ [Accessed 12th February 2020].
- Field T. S. et al. (July 2008). Costs associated with developing and implementing a
computerized clinical decision support system for medication dosing for patients with renal
insufficiency in the long-term care setting. Journal of the American Medical Informatics
Association. J Am Med Inform Assoc, 15(4), 466–472. Available from: https://www.ncbi.
nlm.nih.gov/pmc/articles/PMC2442257/ [Accessed 18th February 2020].
- Fifth Third Bank. (n.d.). Cobots Could Help Boost Manufacturers’ Productivity and Profits.
Available from: https://www.53.com/content/fifth-third/en/commercial-banking/resource-
center/acting-on-your-industry/robots-boost-manufacturer-productivity.html [Accessed 7th
February 2020].
- Forbes. (February 2018), The future of marketing mix optimization is here. https://www.
forbes.com/sites/baininsights/2018/02/28/the-future-of-marketing-mix-optimization-is-
here/#51c58cce1387, [Accessed 5th February 2020]
- Forger, G. (n.d.). Smart Glasses: The latest weapon against downtime. Available from:
https://www.mmh.com/article/smart_glasses_the_latest_weapon_against_downtime
[Accessed 7th February 2020].
- Freed J.F. (December 2018) Efficient Yielding Strategies Drive Big Gains for Coast Hotels.
Duetto. Available from: https://www.duettocloud.com/library/efficient-yielding-strategies-
lead-to-big-gains-for-coast-hotels [Accessed 5th February 2020]
- Freshdesk. (n.d.) Grofers elevates real-time customer experience for its users with
Freshdesk. Available from: https://freshdesk.com/resources/case-study/grofers [Accessed
5th February 2020]
- Freshworks. (n.d. a) How UK-based family business, Finders International improved
customer targeting with Freshsales CRM. Available from: https://www.freshworks.com/
freshsales-crm/lp/resources/case-study/finders-international/ [Accessed 7th February
2020].
- Freshworks. (n.d. b) You’re in good company. Available from: https://www.freshworks.com/
customers/ [Accessed 7th February 2020].
- Fujita Kanko Inc. (April 2017). Hotels Implement Multilingual AI Chatbot. PRNewswire.com.
Available from: https://www.prnewswire.com/news-releases/fujita-kankos-japan-based-
hotels-implement-multilingual-ai-chatbot-300564286.html [Accessed 20th February 2020].
- Gandorfer M. (April 2006) Bewertung von Precision Farming dargestellt am Beispiel der
teilflächenspezifischen Stickstoffdüngung. (Doctoral dissertation, Technische Universität
München). Available from: https://mediatum.ub.tum.de/doc/603702/603702.pdf).
[Accessed 26th February].

78
- GDPR.EU. (n.d.) Complete guide to GDPR compliance. Available from: https://gdpr.eu/
[Accessed 6th February 2020].
- Genesys. (n.d. a) PureCloud pricing. Available from: https://www.genesys.com/pricing
[Accessed 5th February 2020].
- Genesys. (n.d. b) Taking customer experience to new levels by differentiating service
through digital, mobile and social. Available from: https://www.genesys.com/customer-
stories/vodafone [Accessed 5th February 2020]
- Geoconcept (2014). Témoignage Client: Franprix Leader Price. Available from: https://
fr.geoconcept.com/entreprise/temoignage-client-franprix-leader-price [Accessed 7th
February 2020].
- Geoconcept (2019). Messer. Available from: https://fr.geoconcept.com/cas-clients/messer
[Accessed 7th February 2020].
- Gerhards R. et al. (1997) Site Specific Weed Control in Winter Wheat. In Journal of
Agronomy & Crop Science 178, p.219-225.
- Github (n.d.) GummiArm. Available from: http://mstoelen.github.io/GummiArm/ [Accessed
19th February 2020].
- Graamans, L., et al. (February 2017). Plant factories versus greenhouses: Comparison of
resource use efficiency. Agricultural Systems. 160.10.1016/j.agsy.2017.11.003.
- Green H.M. et al. (1997) Precision Management of Spatially Variable Weeds. In STAFFORD,
J.V. (Ed.): Precision Agriculture 1997, Volume II: Technology, IT and Management, Oxford,
p.983-989.

- Grossman et al. (2011) Technology-Assisted Review in E-Discovery Can Be More Effective

and More Efficient Than Exhaustive Manual Review, XVII RICH. J.L. & TECH. 11 (2011) at
23-24).
- Guerriero, F. et al. (January 2012) Revenue management policies for the truck rental
industry. Transportation Research Part E: Logistics and Transportation Review, 48(1), 202-
214
- Harris D. (1997) Risk Management in Precision Farming. In STAFFORD, J.V. (Ed.):
Precision Agriculture 1997, Volume II: Technology, IT and Management, Oxford, p.949-
956.
- Harvest Croo (n.d.) Technology Highlights. Available from: https://harvestcroo.com/
about/#technology-highlights [Accessed 19th February 2020].
- Harvey R. (February 2017) How the Internet of Things will revolutionise agriculture. Bosch
ConnectedWorld Blog. Available from: https://blog.bosch-si.com/projects/how-the-
internet-of-things-will-revolutionise-agriculture/ [Accessed 19th February 2020].
- Health IT Analytics. (December 2017). Understanding the basics of clinical decision
support systems. Available from: https://healthitanalytics.com/features/understanding-
the-basics-of-clinical-decision-support-systems [Accessed 13th February 2020].
- Healthy.io (n.d.) Homepage. Available from: https://healthy.io/ [Accessed 12th February

79
 2020].
- Helfand, M. et al. (May 2016) Technology Assessment: Early Sense for Monitoring
Vital Signs in Hospitalized Patients. Available from: https://www.hsrd.research.va.gov/
publications/esp/EarlySense.pdf [Accessed 4th February 2020].
- HERE. (n.d.) Fleet Optimization, Maximizing Fleet Utilization, HERE. Available from: https://
www.here.com/fleet-utilization [Accessed 4th February 2020].
- Hobbs, H. (August 2015) Caterpillar helps Black Thunder mine enhance dozer productivity.
Available from: https://www.worldcoal.com/mining/12082015/caterpillar-helps-black-
thunder-mine-enhance-dozer-productivity-2176/ [Accessed 19th February 2020].
- Hotelchamp. (n.d.) Case study New Century Hotel. Available from: https://www.
hotelchamp.com/wp-content/uploads/2017/05/HC_Case_NewCentury-1.pdf [Accessed
5th February 2020].
- Hypatos. (n.d.) Homepage. Available from: https://hypatos.ai/en/pricing [Accessed 19th
February 2020].
- Iagri. (n.d.) Homepage. Available from: https://www.iagri.com/ [Accessed 19th February
2020].
- Japp. (2019) Privacy Tech Vendor Report v.3.2. Available from: https://iapp.org/media/
pdf/resource_center/2019TechVendorReport.pdf [Accessed 6th February 2020].
- IBM (n.d.) Build AI chatbots employees want to talk to. Available from: https://www.ibm.
com/thought-leadership/institute-business-value/report/hrchatbots# [Accessed 18th
February 2020].
- IBM. (n.d.) Artificial intelligence for a smarter kind of cybersecurity. Available from:
https://www.ibm.com/security/artificial-intelligence [Accessed 6th February 2020].
- IBM. (n.d.) Smart Manufacturing with IBM Production Optimization. Available from:
https://www.ibm.com/services/process/iot/production-optimization [Accessed 5th
February 2020].
- Ideal. (n.d. a) AI for Recruiting: News, Tips, and Trends. Available from: https://ideal.com/
predictive-recruitment-analytics/ [Accessed 6th February 2020].
- Ideal. (n.d. b) Customers. Available from: https://ideal.com/customer/indigo-books-music/
[Accessed 6th February 2020].
- IHS. (February 2020) Markit Comparative Industry Service. Available from: https://
ihsmarkit.com/products/global-industry-forecasts-analysis.html [Accessed 28th February
2020]
- Impactvi. (n.d.) Non-invasive real-time avocado dry matter content. Available from: https://
www.impactvi.com/product/avocado-dry-matter/ [Accessed 19th February 2020].
- InnospeXion. (n.d.) Latest news. Available from: https://innospexion.dk/2019/07/31/
avocado-ripeness-inspection-xray/ [Accessed 19th February 2020].
- Intel. (n.d.) Intel-Powered AI Helps Optimize Crop Yields. Available from: https://www.intel.
com/content/www/us/en/big-data/article/agriculture-harvests-big-data.html [Accessed

80
 19th February 2020].
- International Finance Corporation. (n.d.) Impact of Efficient Irrigation Technology on
Small Farmers. World Bank Group. Available from: https://www.ifc.org/wps/wcm/
connect/1f630d98-dabc-41e4-9650-b8809d620664/Impact+of+Efficient+Irrigation+T
echnology+on+Small+Farmers+-+IFC+Brochure.pdf?MOD=AJPERES&CVID=lKbEzwG
[Accessed 5th February 2020].
- Inviarobotics.com (n.d.). inVia Robotics: Autonomous Warehouse Robots, Picking Robot.
Available from: https://www.inviarobotics.com/ [Accessed 12th February 2020].
- IoT Business News. (February 2019) Is Telematics Ultimately Worth the Cost? Available
from: https://iotbusinessnews.com/2019/02/21/17655-is-telematics-ultimately-worth-
the-cost/ [Accessed 5th February 2020].
- Ironscales (n.d. a) Ironscales stops polymorphic phishing attack. Available from: https://
get.ironscales.com/rs/954-HBD-655/images/ironscales_usecase_construction.pdf
[Accessed 12th February 2020].
- Ironscales (n.d. b) Gamified, Personalized Phishing Simulation & Training. Available from:
https://ironscales.com/anti-phishing-solutions/phishing-simulation/ [Accessed 12th
February 2020].
- Ironscales (n.d. c) Ironscales provides robust email security for midsized PR agency.
Available from: https://get.ironscales.com/rs/954-HBD-655/images/arpr_case_study.pdf
[Accessed 12th February 2020].
- Itemize. (n.d.) Accurate Invoice Processing. Available from: https://itemize.com/api/
automated-invoice-processing/ [Accessed 5th February 2020]
- Khaki, S. and Wang, L. (May 2019) Crop Yield Prediction Using Deep Neural Networks.
Industrial and Manufacturing Systems Engineering, Iowa State University. Available from:
https://www.frontiersin.org/articles/10.3389/fpls.2019.00621/full [Accessed 28th February
2020].
- Khan, O. (November 2017). The Impact of Machine Learning & AI on Geomatics. Available
at: https://gogeomatics.ca/the-impact-of-machine-learning-ai-on-geomatics/ [Accessed
5th February 2020].
- Khokhar, T. (March 2017) Chart: Globally, 70% of Freshwater is Used for Agriculture. World
Bank Blogs. Available from: https://blogs.worldbank.org/opendata/chart-globally-70-
freshwater-used-agriculture [Accessed 5th February 2020].
- Kira (n.d. a) Accurate Due Diligence, Done in Minutes. Available from: https://kirasystems.
com/how-it-works/due-diligence/ [Accessed 19th February 2020].
- Kira b (n.d. b) Fredrikson Law increases due diligence efficiency Firm-wide by over 20%.
Available from: https://kirasystems.com/resources/case-studies/fredrikson/ [Accessed
19th February 2020].
- Kira Systems (n.d. c) Case Study: Integreon leverages Kira to help a global ride-sharing
company reduce overall contract review time by 40%. Available from: https://kirasystems.

81
 com/resources/case-studies/integreon/ [Accessed 18th February 2020].
- Kira Systems (n.d. d) It’s how you know what’s in your contracts. Available from: https://
kirasystems.com/ [Accessed 18th February 2020].
- Kite-Powell, J. (March 2019) This Autonomous Agricultural Robot Creates Maps And
Routes Of Fields In Real Time. Forbes.com. Available from: https://www.forbes.com/sites/
jenniferhicks/2019/03/18/this-autonomous-agricultural-robot-creates-maps-and-routes-
of-fields-in-real-time/ [Accessed 19th February 2020].
- Klahr, R. et al. (April 2017). Cyber security breaches survey 2017: main report. Available
from: https://assets.publishing.service.gov.uk/government/uploads/system/uploads/
attachment_data/file/609186/Cyber_Security_Breaches_Survey_2017_main_report_
PUBLIC.pdf [Accessed 12th February 2020].
- Kounalakis, T. et al. (2017). Vision System for Robotized Weed Recognition in Crops and
Grasslands. International Conference on Computer Vision Systems.pp. 485-498.
- Legal.thomsonreuters.com. (n.d.) Artificial Intelligence and Corporate Legal Departments.
Available from: https://legal.thomsonreuters.com/en/insights/articles/artificial-
intelligence-ai-report [Accessed 13th February 2020].
- Leifels, A. (November 2018). Many job openings, few job seekers: SMEs expect a shortage
of skilled workers. KfW. Available from: https://www.kfw.de/PDF/Download-Center/
Konzernthemen/Research/PDF-Dokumente-Fokus-Volkswirtschaft/Fokus-englische-
Dateien/Fokus-2018-EN/Fokus-Nr.-232-November-2018-Skilled-worker-shortage_EN.pdf
[Accessed 19th February 2020].
- LfL Institut für Geospatial big data handling theory and methods: a review and research
challenges.Pflanzenschutz. (n.d.) Rechtliche Vorschriften. Available from: https://www.lfl.
bayern.de/ips/recht/index.php [Accessed 3rd February 2020].
- Li, S., et al. (2016) ISPRS J Photogramm Remote Sens. 115:119-33.
- logistics.dhl. (May 2019) DHL supply chain deploys latest version of smart glasses
worldwide. Available from: https://www.logistics.dhl/global-en/home/press/press-
archive/2019/dhl-supply-chain-deploys-latest-version-of-smart-glasses-worldwide.html
[Accessed 12th February 2020].
- Loidl, M. et al. (2016). GIS and Transport Modeling – Strengthening the Spatial
Perspective. ISPRS International Journal of Geo-Information. 5(6), p. 84.
- Lowe, H. (n.d.). Warehouse Automation. Available from: https://www.selecthub.com/
warehouse-management/building-automated-warehouse-system [Accessed 12th
February 2020].
- Madden-Jones, K. (December 2019) Revenue Analytics Unveils FareVantage™, Next-
Generation Revenue Management for Passenger Rail. Available from: https://www.
businesswire.com/news/home/20191202005072/en/Revenue-Analytics-Unveils-
FareVantage%E2%84%A2-Next-Generation-Revenue-Management [Accessed 4th February
2020].

82
- Manning, J. (February 2018) Fleet telematics must comply with EU’s new data protection
rules. Available from: https://www.fleeteurope.com/en/new-energies/europe/analysis/
fleet-telematics-must-comply-eus-new-data-protection-rules?a=JMA06&t%5B0%5D=GDP
R&t%5B1%5D=EU&curl=1 [Accessed 5th February 2020].
- Maurer, M. (August 2019). Airbus Harnessing AI in Bid to Save Millions on Finance Tasks.
The Wall Street Journal. Available from: https://www.wsj.com/articles/airbus-harnessing-
ai-in-bid-to-save-millions-on-finance-tasks-11566207002 [Accessed 4th February 2020].
- McKinsey. (April 2017) Smartening Up with AI – what’s in it for the industrial
sector. Available from: https://www.mckinsey.com/~/media/McKinsey/Industries/
Semiconductors/Our%20Insights/Smartening%20up%20with%20artificial%20intelligence/
Smartening-up-with-artificial-intelligence.ashx [Accessed 5th February 2020].
- Medium. (June 2019) Cookies, ITP and how it affects your privacy. Available from: https://
medium.com/swlh/cookies-itp-and-how-it-affects-your-privacy-7ad39c9de46 [Accessed
5th February 2020].
- Mehta, S. S. and Burks, T. F. (2014). Vision-based control of robotic manipulator for citrus
harvesting. Computers and Electronics in Agriculture. 102. 146–158.
- MGI. (January 2015) Global growth: Can productivity save the day in an aging world?.
Available from: https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/
Employment%20and%20Growth/Can%20long%20term%20global%20growth%20be%20
saved/MGI_Global_growth_Full_report_February_2015pdf.ashx [Accessed 12th February
2020].
- MGI. (June 2015) A window of opportunity for Europe. Available from: https://www.
mckinsey.com/featured-insights/europe/a-window-of-opportunity-for-europe [Accessed
12th February 2020].
- MGI. (June 2016) Digital Europe: Pushing the frontier, capturing the benefits. Available
from: https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/
McKinsey%20Digital/Our%20Insights/Digital%20Europe%20Pushing%20the%20
frontier%20capturing%20the%20benefits/Digital-Europe-Full-report-June-2016.ashx
[Accessed 12th February 2020].
- MGI. (January 2017) A future that works: automation, employment, and productivity
Available from: https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/
Digital%20Disruption/Harnessing%20automation%20for%20a%20future%20that%20works/
MGI-A-future-that-works_Full-report.ashx [Accessed 28th February 2020].
- MGI. (February 2017) Reinventing construction: a route to higher productivity Available
from: https://www.mckinsey.com/~/media/McKinsey/Industries/Capital%20Projects%20
and%20Infrastructure/Our%20Insights/Reinventing%20construction%20through%20
a%20productivity%20revolution/MGI-Reinventing-Construction-Executive-summary.ashx
[Accessed 28th February 2020].
- MGI. (October 2017) Shaping the future of work in Europe’s digital front-runners. Available

83
 from: https://www.mckinsey.com/~/media/mckinsey/featured%20insights/europe/
shaping%20the%20future%20of%20work%20in%20europes%20nine%20digital%20
front%20runner%20countries/shaping-the-future-of-work-in-europes-digital-front-
runners.ashx [Accessed 12th February 2020].
- MGI. (December 2017) Jobs lost, jobs gained: Workforce transition in a time of
automation. Available from: https://www.mckinsey.com/featured-insights/future-of-work/
jobs-lost-jobs-gained-what-the-future-of-work-will-mean-for-jobs-skills-and-wages [Accessed
28th February]
- MGI. (February 2018) Solving the productivity puzzle: The role of demand and the promise
of digitization. Available from: https://www.mckinsey.com/~/media/mckinsey/featured%20
insights/meeting%20societys%20expectations/solving%20the%20productivity%20puzzle/
mg-solving-the-productivity-puzzle--report-february-2018.ashx [Accessed 12th February
2020].
- MGI. (April 2018) Notes from the AI Frontier: Applications and value of deep learning.
Available from https://www.mckinsey.com/featured-insights/artificial-intelligence/notes-
from-the-ai-frontier-applications-and-value-of-deep-learning [Accessed 5th February
2020].
- MGI. (September 2018) Notes from the AI frontier: Modeling the impact of AI on the world
economy. Available from: https://www.mckinsey.com/~/media/McKinsey/Featured%20
Insights/Artificial%20Intelligence/Notes%20from%20the%20frontier%20Modeling%20
the%20impact%20of%20AI%20on%20the%20world%20economy/MGI-Notes-from-the-
AI-frontier-Modeling-the-impact-of-AI-on-the-world-economy-September-2018.ashx
[Accessed 12th February 2020].
- MGI. (October 2018) Superstars. The dynamics of firms, sectors, and cities leading
the global economy. Available from: https://www.mckinsey.com/~/media/mckinsey/
featured%20insights/innovation/superstars%20the%20dynamics%20of%20firms%20
sectors%20and%20cities%20leading%20the%20global%20economy/mgi_superstars_
discussion%20paper_oct%202018-v2.ashx [Accessed 12th February 2020].
- Michel, R. (November 2018) Insight into smart glasses in warehouses and distribution
centers. Available from: https://www.supplychain247.com/article/insight_into_smart_
glasses_utilization_in_warehouse_and_distribution_center/Augmented_Reality [Accessed
12th February 2020].
- Mileva, G. (February 2019) How Augmented Reality Is Redefining the Fashion Industry.
Available from: https://arpost.co/2019/02/11/how-augmented-reality-is-redefining-the-
fashion-industry/ [Accessed 10th February 2020].
- Miller, P. J. et al. (2016) Case Study: Implementing Early Detection of Patient Deterioration
in Medical and Surgical Units. Biomedical Instrumentation & Technology. Vol. 50, No.6, pp.
439-446.
- Miller, J. (December 2019) What are the costs associated with vehicle telematics?

84
 Available from: https://keeptruckin.com/blog/costs-associated-with-vehicle-telematics/
[Accessed 5th February 2020].
- Mobile-industrial-robots.com. (n.d.) 3 MiR1000 optimize warehouse internal logistics
at ICM. Available from: https://www.mobile-industrial-robots.com/en/resources/case-
studies/3-mir1000-amrs-optimize-warehouse-logistics-at-icm/ [Accessed 12th February
2020].
- Möbius, J. (April 2016) Wasser nach Bedarf. Agrarmanager. Available from: https://
www.beregnungsanlagen-pumpentechnik.de/images/news/agrarmanager_Beregnung.
pdf_%C3%BCberarbeitet.pdf [Accessed 5th February 2020].
- Modern Machine Shop. (January 2019) What This Seat Bracket Says About the Future of
Automotive Manufacturing. Available from: https://www.mmsonline.com/blog/post/what-
this-seat-bracket-says-about-the-future-of-automotive-manufacturing [Accessed 10th
February 2020].
- Morozova, A. (n.d.) From DIY to Deep Learning: 5 Types of Chatbots and How Much Will It
Cost to Develop Them. Available from: https://jasoren.com/from-diy-to-deep-learning-5-
types-of-chatbots-and-how-much-will-it-cost-to-develop-them/ [Accessed 5th February
2020].
- Mosqueda, M. R. P. et al. (2010). Predicting the economics of X-ray inspection technology
in sweet onion packinghouses using simulation modelling. Biosystems Engineering. 105
(1). 139–147.
- Movex. (n.d. a) Fleet management software. Available from: https://www.movex.ai/fleet-
management-software/ [Accessed 5th February 2020].
- Movex. (n.d. b) Pricing. Available from: https://www.movex.ai/fleet-management-
software/pricing [Accessed 5th February 2020].
- Muro, M. and Andes, S. (June 2015) Robots seem to be improving productivity, not costing
jobs. Harvard Business Review. Available from: https://hbr.org/2015/06/robots-seem-to-
be-improving-productivity-not-costing-jobs [Accessed 30th January 2020].
- Nandi, C. S. et al. (2014). A machine vision-based maturity prediction system for sorting
of harvested mangoes. IEEE Transactions on Instrumentation and Measurement. 63(7).
1722–1730.
- Naus, T. (August 2018). Is vertical farming really sustainable? Available from: https://
www.eitfood.eu/blog/post/is-vertical-farming-really-sustainable [Accessed 10th February
2020].
- Neuronews (September 2015) VA Spinal Cord Injury Center demonstrates
clinical improvement with EarlySense monitoring system. Available from: https://
neuronewsinternational.com/va-spinal-cord-injury-center-demonstrates-clinical-
improvement-with-earlysense-monitoring-system/ [Accessed 26th February 2020]
- New Equipment Digest. (March 2018) What Generative Design Is and Why It’s the Future
of Manufacturing. Available from: https://www.newequipment.com/research-and-

85
 development/article/22059780/what-generative-design-is-and-why-its-the-future-of-
manufacturing [Accessed 10th February 2020].
- Nordmeyer, H. et al. (1997) Patchy weed control as an approach in precision farming.
Precision agriculture papers presented at the first European Conference on Precision
Agriculture, Warwick University Conference Centre.
- Nuance Communications. (n.d.) Dragon® Medical One monthly subscription –- 1 year
term. Available from: https://www.dictation.cloud/shop/power [Accessed 20th February
2020].
- Nuance Communications. (October 2018) AI-powered speech recognition eases clinical
documentation burden. Available from: https://www.nuance.com/content/dam/nuance/
en_us/collateral/healthcare/case-study/cs-dragon-medical-one-embedded-in-epic-allina-
health-en-us.pdf [Accessed 20th February 2020].
- Octinon. (n.d.) Rubion. Available from: http://octinion.com/products/agricultural-robotics/
rubion [Accessed 19th February 2020].
- OECD. (2015) Frontier firms, technology diffusion and public policy: micro evidence from
OECD countries. Available from: http://www.oecd.org/economy/growth/Frontier-Firms-
Technology-Diffusion-and-Public-Policy-Micro-Evidence-from-OECD-Countries.pdf
[Accessed 12th February 2020].
- OECD. (February 2018) Strengthening SMEs and entrepreneurship for productivity
and inclusive growth. Available from: https://www.oecd.org/cfe/smes/ministerial/
documents/2018-SME-Ministerial-Conference-Key-Issues.pdf [Accessed 12th February
2020].
- Optimove. (n.d.) Optimove helps Sisal Boost LTV and Revenues by Becoming More
Customer Centric. Available from: https://www.optimove.com/resources/customer-
success-stories/download-sisal-case-study [Accessed 12th February 2020].
- Orchard, R. (April 2018) Easy pickings? How robot farm hands could revolutionise
agriculture. Slow journalism.com. Available from: https://www.slow-journalism.
com/delayed-gratification-magazine/easy-pickings-how-robot-farm-hands-could-
revolutionise-agriculture [Accessed 19th February 2020].
- Oxford Insights. (2019) Government Artificial Intelligence Readiness Index 2019. Available
from: https://www.oxfordinsights.com/ai-readiness2019 [Accessed 30th January 2020].
- Panasonic. (May 2018) Introducing AI-equipped Tomato Harvesting Robots to
Farms May Help to Create Jobs. Available from: https://news.panasonic.com/global/
stories/2018/57801.html [Accessed 19th February 2020].
- Partel, V. et al. (2019). Development and evaluation of a low-cost and smart technology for
precision weed management utilizing artificial intelligence. Computers and electronics in
agriculture, 157, 339-350.
- Paycom. (July 2020) Paycom’s Direct Data Exchange Update Reflects Latest EY Findings.
Available from: https://investors.paycom.com/press-releases/press-release-details/2020/

86
 Paycoms-Direct-Data-Exchange-Update-Reflects-Latest-EY-Findings/default.aspx#.
XkvsRGhKhPY [Accessed 18th February 2020].
- PayScale. (n.d.) Average Data Scientist Salary in Germany. Available from: https://www.
payscale.com/research/DE/Job=Data_Scientist/Salary [Accessed 7th February 2020].
- Pearson, K. et al. (2010) Evidence-based Falls Prevention in Critical
Access Hospitals. University of Southern Maine. Available from: https://
digitalcommons.usm.maine.edu/cgi/viewcontent.cgi?referer=https://www.google.
com/&httpsredir=1&article=1014&context=rural_hospitals [Accessed 12th February 2020].
- Peck, A. (March 2019) FDA Approves Smartphone-based Urinalysis Test Kit for At-Home
Use That Matches Quality of Clinical Laboratory Tests. Darkdaily.com. Available from:
https://www.darkdaily.com/fda-approves-smartphone-based-urinalysis-test-kit-for-at-
home-use-that-matches-quality-of-clinical-laboratory-tests/ [Accessed 12th February
2020].
- Pellenc. (n.d.) Sorting berry by berry: Selectiv’ Process Vision 2. Available from: https://
pellenc.com/agri/produits/sorting-berry-by-berry-selectiv-process-vision-2/?lang=en
[Accessed 19th February 2020].
- Performance livestock analytics (n.d.) Performance Beef. Available from: https://www.
performancelivestockanalytics.com/performance-beef/ [Accessed 19th February 2020].
- Pessl Instruments GmbH. (n.d.) Water Management. Available from: http://metos.at/
water-management/ [Accessed 4th February 2020].
- Plastima. (January 2020) Estimated Costs of Crop Production in Iowa - 2020. Iowa State
University. Available from: https://store.extension.iastate.edu/Product/fm1712-pdf
[Accessed 4th February 2020].
- Porcellana, P. (March 2016) What the EU’s latest data protection law means for chatbot
makers and marketers. Available from: https://venturebeat.com/2018/03/16/what-the-
eus-latest-data-protection-law-means-for-chatbot-makers-and-marketers/ [Accessed 5th
February 2020].
- Porcellana, P. (March 2016). What the EU’s latest data protection law means for chatbot
makers and marketers. VentureBeat. Available from: https://venturebeat.com/2018/03/16/
what-the-eus-latest-data-protection-law-means-for-chatbot-makers-and-marketers/
[Accessed 5th February 2020].
- Postscapes.com. (n.d.) Smart Glasses for Warehouses. Available from: https://www.
postscapes.com/smart-glasses-for-warehouses-smartpick/ [Accessed 12th February
2020].
- Prasad, G. et al. (June 2019) Agriculture Commodity Arrival Prediction using Remote
Sensing Data: Insights and Beyond. Microsoft India Dev. Centre (IDC). Available from:
https://arxiv.org/pdf/1906.07573.pdf [Accessed 28th February 2020].
- Prefixbox. (n.d.) Praktiker Hungary Ecommercial Digital Transformation Story. Available
from: https://www.prefixbox.com/en-us/AboutUs/CaseStudyPraktiker [Accessed 13th

87
 February 2020].
- Priyanka, T. (March 2019) Agricultural Crop Yield Prediction Using Artificial Intelligence and
Satellite Imagery. Eurasian Journal of Analytical Chemistry. Available from: [Accessed 28th
February 2020]. [Accessed 28th February 2020].
- PRNewswire. (February 2020) Highlights Accuracy of EarlySense System for Continuously
Monitoring High-risk Hospital Patients. Available from: https://www.prnewswire.com/
news-releases/clinical-validation-study-highlights-accuracy-of-earlysense-system-for-
continuously-monitoring-high-risk-hospital-patients-300998294.html [Accessed 5th
February 2020].
- PROS. (n.d.) Car Rental Company Generates Millions in Additional Revenue. Available from:
https://resources.pros.com/automotive/car-rental-company-generates-millions [Accessed
26th February 2020]
- Prospera. (n.d.) Transforming the way food is grown with Data and Artificial Intelligence.
Available from: https://home.prospera.ag/ [Accessed 19th February 2020].
- Qventus. (n.d. a) Success Story - New York Presbyterian. Available from: https://qventus.
com/success-story-new-york-presbyterian/ [Accessed 5th February 2020]
- Qventus. (n.d. b) Success Story – SCL Health Lutheran. Available from: https://qventus.
com/success-story-scl-health-lutheran/ [Accessed 5th February 2020].
- RASberry. (n.d.) RASberry - Robotics and Automation Systems for berry production.
Available from: https://rasberryproject.com/ [Accessed 19th February 2020].
- Rayo, E. (November 2019) AI in Law and Legal Practice – A Comprehensive View of 35
Current Applications. Emerj. Available from: https://emerj.com/ai-sector-overviews/ai-in-
law-legal-practice-current-applications/ [Accessed 13th February 2020].
- Redaktion agrarheute. (May 2016) Teilflächenspezifische Beregnung: Ein Praxisbeispiel.
Available from: https://www.agrarheute.com/pflanze/getreide/teilflaechenspezifische-
beregnung-praxisbeispiel-522944 [Accessed 5th February 2020].
- Relativity6. (n.d.) Case Study: Reducing Churn For An Insurance Carrier With Artificial
Intelligence. Available from: https://www.relativity6.com/case-studies/reducing-churn-
insurance-carrier-artificial-intelligence [Accessed 5th February 2020].
- Revenue Analytics. (n.d.) Major Commuter & Leisure Rail Company Saves 100s of Hours Per
Week. Available from: https://www.revenueanalytics.com/rail/resources/rail-co-saves-100s-
of-hours [Accessed 26th February 2020]
- Rieger. (October 2018) Alpha Omega Winery Tests New “Alien” Grape Sorting
Robot. WineBusiness.com. Available from: https://www.researchgate.net/
publication/332653186_The_Impact_of_Swarm_Robotics_on_Arable_Farm_Size_and_
Structure_in_the_UK [Accessed 19th February 2020].
- Robert Bosch GmbH. (2018) IoT Based Smart Irrigation System. Available from: https://
www.bosch.com/stories/iot-based-smart-irrigation-system/ [Accessed 5th February
2020].

88
- Robinson, A. et al. (May 2019) Robots Take the Wheel as Autonomous Farm
Machines Hit Fields. Bloomberg. Available from: https://www.bloomberg.com/news/
articles/2019-05-15/robots-take-the-wheel-as-autonomous-farm-machines-hit-the-field
[Accessed 19th February 2020].
- Robotics and Automation. (July 2018) 4 reasons why collaborative robotics is the
answer to SME automation. Available from: https://roboticsandautomationnews.
com/2018/07/08/4-reasons-why-collaborative-robotics-is-the-answer-to-sme-
automation/18140/ [Accessed 19th February 2020].
- Robotics Online. (May 2018) The Latest Technological Innovations in Autonomous
Mobile Robots. Available from: https://www.robotics.org/blog-article.cfm/The-Latest-
Technological-Innovations-in-Autonomous-Mobile-Robots/97 [Accessed 12th February
2020].
- Romaine, E. (2019). Automated Storage & Retrieval System (AS/RS) – Types & Uses.
Available from: https://www.conveyco.com/automated-storage-and-retrieval-types/
[Accessed 12th February 2020].
- Rossum. (n.d.) Data Capture with Rossum. Available from: https://rossum.ai/data-
capture#success_stories [Accessed 19th February 2020].
- Rubenstone, J. (June 2019). Mortenson Reaches Agreement To Deploy Fully Autonomous
Equipment. Available from: https://www.enr.com/articles/46996-mortenson-reaches-
agreement-to-deploy-fully-autonomous-equipment [Accessed 19th February 2020].
- Ruekkasaem, L. and Sasananan, M. (July 2018) Forecasting agricultural products prices
using time series methods for crop planning. Department of Industrial Engineering,
Faculty of Engineering, Thammasat University. Available from: http://www.iaeme.com/
MasterAdmin/UploadFolder/IJMET_09_07_102/IJMET_09_07_102.pdf [Accessed 28th
February 2020].
- SAGA. (n.d.) Homepage. Available from: http://laral.istc.cnr.it/saga/ [Accessed 19th
February 2020].
- Saha, V. (February 2020) What is Warehouse Automation? Definition, Types & More.
Available from: https://www.camcode.com/asset-tags/what-is-warehouse-automation/
[Accessed 12th February 2020].
- Saha, V. (January 2020 ) Benefits of Warehouse Automation: Advice from 25 Experts.
Available from: https://www.camcode.com/asset-tags/benefits-of-warehouse-automation/
[Accessed 12th February 2020].
- Sampson, J. (December 2019). MAN Energy Solutions launches pilot phase of assisted-
reality application. Gasworld. Available from: https://www.gasworld.com/man-energy-
launches-pilot-phase-of-application/2018236.article [Accessed 19th February 2020].
- Sato, K. (August 2016) How a Japanese cucumber farmer is using deep learning and
TensorFlow. Google Could Platform. Available from: https://cloud.google.com/blog/
products/gcp/how-a-japanese-cucumber-farmer-is-using-deep-learning-and-tensorflow

89
 [Accessed 19th February 2020].
- Scanwell/Lemonaid (n.d.). Homepage. Available from: https://www.lemonaidhealth.com/
services/urinary-tract-infection [Accessed 12th February 2020].
- Scharfenberger Maschinenbau. (n.d.) Euroselect Optical Sorter. Available from: https://
www.scharfenberger.de/index.php/en/products/processing-systems/optical-sorter.html
[Accessed 19th February 2020] .
- Schmid, J. (n.d.) What is the Cost of Telematics? Available from: https://www.gpsinsight.
com/blog/what-is-the-cost-of-telematics/ [Accessed 5th February 2020].
- Searchnode. (n.d.) Decathlon.pl increases site search conversion by 33%. Available from:
https://searchnode.com/wp-content/uploads/2019/05/Decathlon.pl-Site-Search-Case-
Study.pdf [Accessed 13th February 2020].
- SeeUnity. (n.d.) Partner: Kira Systems. Available from: https://seeunity.com/kira_systems/
[Accessed 18th February 2020].
- Serket-tech. (n.d.) Products. Available from: https://www.serket-tech.com/Products
[Accessed 19th February 2020]
- Shibuya Seiki. (n.d.) Product information. Available from: https://www.shibuya-sss.co.jp/
sss_e/product/grading.html [Accessed 19th February 2020].
- Shieber, J. (November 2019) Scanwell Health launches smartphone tests for UTIs in
partnership with Lemonaid Health. Techcrunch.com. Available from: https://techcrunch.
com/2019/11/05/scanwell-health-launches-smartphone-tests-for-utis-in-partnership-
with-lemonaid-health/?guccounter=1 [Accessed 12th February 2020].
- Silvon.com. (n.d.) Guidelines for Estimating the Payback of a Demand Forecasting
Solution. Available from: https://www.silvon.com/blog/demand-forecasting-solution-roi/,
[Accessed 19th February 2020].
- Sinclaire S. (February 2019) UK: AI & Construction Law: An Essential And
Inevitable Partnership - Part 1 Risk And Contract Management. International
Quarterly. Available from: http://www.mondaq.com/uk/x/784712/new+technology/
AI+Construction+Law+An+Essential+And+Inevitable+Partnership [Accessed 18th February
2020].
- Skeletontech. (n.d.) Ultracapacitors for the Rail Industry. Available from: https://www.
skeletontech.com/train-and-rail-industry [Accessed 5th February 2020].
- Skov. (n.d.) Better business and overview. Available from: https://www.skov.com/en
[Accessed 19th February 2020].
- Slight et al. (2014) The return on investment of implementing a continuous monitoring
system in general medical-surgical units. Crit Care Med.;42(8):1862-8.
- Smallbusiness.co.uk (March 2019). Almost a third of small businesses have no cyber
security strategies. Available from: https://smallbusiness.co.uk/cyber-security-
strategies-2547138/ [Accessed 12th February 2020].
- Smallbusiness.co.uk. (March 2019) Almost a third of small businesses have no cyber

90
 security strategies. Available from: https://smallbusiness.co.uk/cyber-security-
strategies-2547138/ [Accessed 12th February 2020].
- Smallrobotcompany. (n.d.) Meet the robots. Available from: https://www.
smallrobotcompany.com/meet-the-robots [Accessed 19th February 2020].
- Smartpick.be. (n.d.) Order Picking with Smart Glasses. Available from: http://www.
smartpick.be/ [Accessed 12th February 2020].
- Smartvid.io. (n.d.) Significantly reduce jobsite risk with the power of human and artificial
intelligence. Available from: https://www.smartvid.io/ [Accessed 12th February 2020].
- Smartvid.io. (n.d. b) Case Study: How Suffolk & Smartvid.io learned to predict and prevent
construction incidents. Available from: https://www.smartvid.io/suffolk-predictive-
analytics-case-study [Accessed 12th February 2020].
- Smartvid.io. (n.d. c) Shawmut design and construction: Using AI to manage project risk
at scale. Available from: https://cdn2.hubspot.net/hubfs/710233/collateral/SmartVid_
IO_Case_Study_Shawmut-Design-and-Construction_FINAL.pdf [Accessed 12th February
2020].
- Südzucker AG. (June 2016) Biorübenanbau ab 2017. Available from: http://www.
engemann-bio.de/images/newsletter-links/S%C3%BCdzucker-Info%20zum%20Bio-
R%C3%BCbenanbau%20ab%202017.pdf [Accessed 3rd February 2020].
- Suki.AI. (September 2019) Plastic surgery specialist reduces timer per patient note by
more than 80% with Suki. Available from: https://resources.suki.ai/doctor/case-study-dr-
ereso-plastic-surgeon [Accessed 20th February 2020].
- Swarm Farm. (n.d.) Our Journey. Available from: https://www.swarmfarm.com/journey/
[Accessed 19th February 2020].
- Sweeper. (n.d.) ICT Robotic Use Cases project in the H2020 programme of the EU.
Available from: http://www.sweeper-robot.eu/ [Accessed 19th February 2020].
- Swinton S.M. and Ahmad M. (1996) Returns to Farmer Investments in Precision
Agriculture Equipment and Services. In ROBERT, P.C., RUST, R.H. and LARSON, W.E. (Eds.):
Proceedings of the Third International Conference on Precision Agriculture, Minneapolis,
p.1009-1018.
- Syte.ai. (n.d. a) i Available from: https://www.syte.ai/camera-search/ [Accessed 14th
February 2020].
- Syte.ai. (n.d. b) Discover How Boohoo Doubles Conversion Rates with Visual Search.
Available from: https://www.syte.ai/boohoo-syte-case-study/ [Accessed 14th February
2020].
- Syte.ai. (n.d. c) Visual AI improves product discovery across Farfetch’s 3,000+ brands.
Available from: https://www.syte.ai/farfetch-syte-case-study/ [Accessed 14th February
2020].
- Telekom Healthcare. (n.d.) iPad statt Aktenordner iMedOne®mobile im
Gemeinschaftskrankenhaus Bonn. Available from:https://www.telekom-

91
 healthcare.com/blob/45026/fd04407f6d7d6ae7b3aea8fa53432776/DL_Referenz_
Gemeinsahftskrankenhaus_Bonn.pdf [Accessed 5th February 2020].
- The Economist. (April 2017) How Germany’s Otto uses artificial intelligence, available from
https://www.economist.com/business/2017/04/12/how-germanys-otto-uses-artificial-
intelligence [Accessed 14th February 2020].
- The Robot Report. (March 2017) 42 companies empowering robots and humans to
work side-by-side. Available from: https://www.therobotreport.com/42-companies-
empowering-robots-and-humans-to-work-side-by-side/ [Accessed 14th February 2020].
- Tipping, A. and Kauschke, P. (2016) Shifting Patterns: The future of the logistics industry.
Available from: https://www.pwc.com/sg/en/publications/assets/future-of-the-logistics-
industry.pdf [Accessed 4th February 2020].
- Tracks for Trucks. (n.d.) We predict your fuel consumption to help you reduce it. Available
from: https://www.tracksfortrucks.com/tracksfortrucks/home/ [Accessed 5th February
2020].
- Trilyo. (n.d.). Discover how to grow direct hotel bookings & maximize revenue. Available
from: https://www.trilyo.com/ [Accessed 17th February 2020].
- Umwelt Bundesamt. (September 2018) Nähr- und Schadstoffe. Available from: https://
www.umweltbundesamt.de/themen/wasser/gewaesser/grundwasser/nutzung-
belastungen/naehr-schadstoffe#belastungen-des-grundwassers [Accessed 3rd February
2020].
- Universal Robots. (January 2016). Meet the “cobots” that are running strong year after
year…after year. Available from: https://blog.universal-robots.com/meet-the-cobots-that-
are-running-strong-year-after-year-after-year [Accessed 4th February 2020].
- Universal Robots. (May 2019) AI and robots: Not what you think. Available from: https://
blog.universal-robots.com/ai-and-robots-not-what-you-think [Accessed 4th February
2020].
- Universal Robots. (n.d.) Vision-guided collaborative robots deliver fast return on
investment in production of firehose valves. Available from https://www.universal-robots.
com/case-stories/task-force-tips/ [Accessed 4th February 2020].
- University of Cambridge. (October 2017) Predicting fruit harvest with drones and artificial
intelligence. Techxplore. Available from:https://techxplore.com/news/2019-10-fruit-
harvest-drones-artificial-intelligence.html [Accessed 4th February 2020].
- Vade Secure. (n.d. a) Anti-Phishing Solution. Available from: https://www.Vade Secure.
com/en/solutions/anti-phishing-2/ [Accessed 4th February 2020].
- Vade Secure. (n.d. b) AI-based Email Security for Office 356. Available from: https://www.
Vade Secure.com/en/dedicated-email-security-for-o365/ [Accessed 4th February 2020].
- Vade Secure. (n.d. c) Gandi optimizes user experience and malicious email protection for
400,000 customers with Vade Secure. Available from: https://www.Vade Secure.com/wp-
content/uploads/Case_Study_Gandi_EN.pdf [Accessed 4th February 2020].

92
- Vectra.ai. (n.d.) Commodities trader finds sweet spot with AI-powered threat detection.
Available from: https://content.vectra.ai/rs/748-MCE-447/images/CaseStudy_2019_
EDFManHoldings_101619.pdf [Accessed 7th February 2020].
- Viz.ai. (n.d.) Viz.ai’s deep learning solution which automatically detects strokes from CTA
scans reducing detection time from 66 to 6 minutes has earned FDA approval. Available
from: https://www.bestpractice.ai/studies/viz_ais_deep_learning_solution_which_
automatically_detects_strokes_from_cta_scans_reducing_detection_time_from_66_to_6_
minutes_has_earned_fda_approval [Accessed 20th February 2020].
- Vogel M. (November 2015) Analysis of Documentation Speed Using Web-Based Medical
Speech Recognition Technology: Randomized Controlled Trial. Journal of Medical Internet
Research. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4642384/
[Accessed 26th February 2020].
- Voice Recognition. (n.d.). Dragon Medical –- official distributorWebsite. Available from:
https://www.voicerecognition.com.au/collections/dragon-medical-practice-edition-4
[Accessed 20th February 2020].
- VR World Tech. (April 2019) Newcastle Surgical Training Centre uses Osso VR. Available
from: https://vrworldtech.com/2019/04/11/newcastle-surgical-training-centre-uses-osso-
vr/ [Accessed 19th February 2020].
- Vuzix (n.d.). Vuzix Remote Assist. Available from: https://www.vuzix.com/vuzix-remote
[Accessed 4th February 2020].
- Vuzix.eu. (n.d.) Vuzix Smart Glasses Solutions: Warehouse Logistics. Available from:
https://www.vuzix.eu/solutions/warehouse [Accessed 12th February 2020].
- Wagner P. (2000) Problems and potential economic impact of precision farming. In
7th ICCTA: Computer Technology in Agricultural Management and Risk Prevention:
Proceedings of the 7th International Congress for Computer Technology in Agriculture
(pp. 241-249).
- Watson, T. (September 2019) Augmented Reality Cost – What is the App Price For 2019?
Available from: https://skywell.software/blog/augmented-reality-cost-what-is-the-app-
price-for-2019/ [Accessed 4th February 2020].
- West Coast University. (April 2019) Maximizing Learning Outcomes by Implementing
Augmented Reality Technology. Available from: https://westcoastuniversity.edu/pulse/
health-e-news/maximizing-learning-outcomes-by-implementing-augmented-reality-
technology.html [Accessed 19th February 2020].
- Wiggers, K. (September 2019) Healthy.io raises $60 million to help patients complete
urine tests on their phone. Venturebeat.com. Available from: https://venturebeat.
com/2019/09/11/healthy-io-raises-60-million-for-smartphone-based-urinalysis-tests/
[Accessed 12th February 2020].
- Wilks D.S. and Wolfe D.W. (1998) Optimal use and economic value of weather forecasts for
lettuce irrigation in a humid climate. Agricultural and forest meteorology, 89(2), 115-129.

93
- Williams, A. (June 2018). How Autonomous Mobile Robots are Changing the Logistics
Landscape. Available from: https://www.roboticsbusinessreview.com/supply-chain/
autonomous-mobile-robots-changing-logistics-landscape/ [Accessed 12th February
2020].
- Wired Workers (n.d.). 7 advantages of using cobots! Available from: https://wiredworkers.
io/advantages-of-cobots/ [Accessed 12th February 2020].
- Wise Athena. (n.d.) Get the right prices every promotional cycle using artificial
intelligence. Available from: https://www.wiseathena.com/ [Accessed 5th February 2020].
- Wossink G.A.A. et al. (2018) The Economics of Agro-Chemicals: An International
Overview of Use Patterns, Technical and Institutional Determinants, Policies
and Perspectives. Routledge. Available from: https://books.google.de/

94
Grow with

95