Integrated Minor Project

Digitalization in demand forecasting using artificial intelligence as a

tool.

Submitted to:-

Archana Gandhi

Submitted by:-

Jayant Kumar
DEMAND FORECASTING

Demand forecasting is a fundamental component of supply chain management in the apparel

industry, influencing decisions related to production, inventory management, and order
fulfillment [1]. With the growing complexity and dynamism of consumer preferences and
market trends, there is a pressing need for more accurate and responsive demand forecasting
methods. Artificial Intelligence (AI) has emerged as a potent tool to address these challenges
[3]. This literature review examines existing research and trends in the application of AI for
enhancing demand forecasting in the apparel sector.AI-powered tools can also help fashion
forecasters spot emerging trends more accurately and quickly [3]. These tools can detect
patterns within large datasets of runway show images, social media posts, search data, and
online and in-store sales data5. This can help forecasters detect emerging trends more
accurately and quickly than traditional qualitative methods such as observing runway shows,
street fashion, and pop culture[3].

Traditional Demand Forecasting

Historically, businesses relied on manual forecasting methods, which often involved expert
judgment and historical sales data [3]. These methods struggled to adapt to the fast-paced
nature of the industry, leading to issues like stockouts and overstocking [3]. These challenges
necessitated a shift towards more data-driven, efficient solutions.

How forecasting was done?

Forecasting was done by using various methods and techniques that relied on historical data,
mathematical models, and human judgment. Some of the common methods and techniques

 Trend analysis: This method used past data to identify the general direction and pattern of
change in the demand over time. For example, a retailer could use trend analysis to forecast
the demand for garments based on the previous year’s sales data [3].
 Moving averages: This method used the average of the most recent data points to smooth
out the fluctuations and noise in the demand data [3] . For example, a restaurant could use
moving averages to forecast the demand for its dishes based on the historical orders data.
 Exponential smoothing: This method used a weighted average of the past and current data
points to give more importance to the recent observations [3]. For example, a smartphone
manufacturer could use exponential smoothing to forecast the demand for its new model
based on the features, price, and reviews of the product.

 Regression analysis: This method used a mathematical equation to describe the

relationship between the demand and one or more independent variables, such as price,
income, advertising, etc. [3]. For example, a clothing retailer could use regression analysis to
estimate the impact of price changes or promotions on the demand for garments.

 Causal models: These models used a system of equations to capture the complex
interactions and feedback among various factors that affect the demand, such as customer
preferences, competitor actions, economic conditions, etc. [3]. For example, a smartphone
manufacturer could use causal models to predict the demand for a specific type of
smartphone or style based on its features and attributes.

 Judgmental methods: These methods used human expertise and intuition to make
subjective estimates or adjustments to the demand forecasts based on qualitative factors,
such as market research, customer feedback, industry trends, etc. For example, a restaurant
could use judgmental methods to forecast the demand for its dishes based on seasonal
variations, promotional activities, and customer feedback [3].
These methods and techniques had some advantages and disadvantages. Some of the
advantages were:

 They were relatively simple and easy to understand and implement.

 They were based on historical data and empirical evidence.
 They could incorporate human knowledge and experience.

Some of the disadvantages were:

 They required constant manual updating of data and adjustments to forecast outputs.
 They were prone to errors and uncertainties due to data quality, model assumptions, or
human bias [3].
 They were not able to handle complex and dynamic situations or learn from new
information or feedback.

Difference b/w Traditional method and Advanced method

Traditional Method Advanced method

 Historical data analysis: Historical sales
data analysis is used by organizations to  Data Analytics and Big Data: The relevant
identify patterns, trends, and seasonality information is gathered and analyzed using
in demand, examining sales figures for advanced analytics and big data technologies,
specific products or services over various encompassing historical sales data, customer
time periods. behavior data, market trends, and social media
 Qualitative Method: Expert opinions and feedback.
qualitative assessments significantly  Machine learning and AI: Machine learning and
influenced the organization, with input AI techniques are increasingly being utilized in
from sales teams, marketing experts, and demand forecasting to automatically identify
industry specialists often supplementing patterns and relationships within data, resulting
quantitative data. in more accurate predictions.
 Market Research: Market research was  Predictive Analytics: Predictive analytics tools
conducted using surveys, focus groups, utilize statistical algorithms and machine
and other methods to gather insights on learning to predict future demand based on
consumer preferences, economic historical and real-time data, enabling
conditions, competitive landscape, and organizations to swiftly adapt to market
external factors affecting demand. changes.
 Sales force input: Sales teams provided  Advance Forecasting models: Organizations use
valuable insights into customer advanced forecasting models like ARIMA and
interactions, market conditions, and Exponential Smoothing to analyze complex
buying behavior, which were incorporated patterns and seasonality in data.
into the forecasting process.  Demand sensing: Demand sensing is a method
 Time series analysis: Statistical that utilizes real-time data and analytics to
techniques like time series analysis were understand and respond to market changes,
used to identify and model historical data especially in industries with rapidly changing
patterns, aiding in forecasting future conditions.
demand based on past performance.  Collaborative planning, forecasting and
 Casual model: Causal models were replenishment(CPFR): CPFR is a supply chain
utilized to examine the correlation management strategy that enhances demand
between demand and various influencing forecast accuracy by involving trading partners
factors such as economic indicators and in forecasting and planning, incorporating inputs
marketing strategies. from various stakeholders.
 Forecasting model: The study utilized  IOT: IOT devices and sensors offer real-time data
various quantitative forecasting models, on product usage and inventory levels, which
including moving averages, exponential are then integrated into forecasting models to
smoothing, and regression analysis, to improve accuracy and responsiveness.
accurately predict and capture demand  Cloud-based solution: Organizations frequently
variations. utilize cloud-based forecasting solutions due to
 Seasonal adjustments: Seasonal their scalability, flexibility, and accessibility,
adjustments were made to account for enabling the processing of large datasets and
regular demand patterns at specific times the deployment of advanced analytics models.
of the year, enabling more accurate  Demand forecasting software’s: Increasingly,
predictions for peak and off-peak periods. demand forecasting software tools are
 ERP system: ERP systems have become becoming more prevalent, often integrated with
increasingly prevalent, integrating sales other enterprise systems, offering user-friendly
and inventory management to provide a interfaces for data analysis and forecast
comprehensive view for forecasting. generation.
 Manual processes: Forecasting was often  Continuous improvement and iterative
a manual process that required significant processes: Organizations utilize agile and
time and effort, often utilizing iterative methods for demand forecasting,
spreadsheet software like Microsoft Excel continuously improving models based on
for calculations and analysis. feedback and incorporating new data as it
becomes available.

AI in Demand Forecasting
AI, particularly machine learning, has gained prominence in demand forecasting [4]. Machine
learning models, such as neural networks, decision trees, and time series analysis, have proven
to be effective tools for predicting demand in the apparel industry [4]. AI-driven solutions could
analyze large datasets and derive insights from various sources, including sales data, social
media trends, and external factors like weather.

Key Components of AI-Driven Demand Forecasting

Data Collection and Preprocessing: The foundation of AI-driven demand forecasting is data.
Companies need to collect, clean, and preprocess various data sources, including historical sales
data, inventory levels, customer reviews, and external factors like seasonality and market trends
[4].

Process of demand forecasting using AI

The process of using AI in demand forecasting can be summarized as follows:

 Collecting and preparing historical sales data and other relevant data sources, such
as weather, holidays, promotions, competitors, etc.
 Choosing and applying appropriate machine learning algorithms, such as regression,
classification, clustering, neural networks, etc., to analyze the data and identify
patterns, trends, and correlations
 Evaluating and validating the accuracy and performance of the models using metrics
such as mean absolute error, root mean square error, mean absolute percentage
error, etc.
 Generating and presenting demand forecasts for different time horizons, such as
short-term, medium-term, and long-term, and for different levels of granularity, such
as product, category, store, region, etc.

 Updating and refining the models and forecasts based on new data and feedback,
and incorporating external factors and uncertainties that may affect the demand.
How AI is being used

Demand forecasting uses artificial intelligence (AI) in many ways. Two common methods for
training AI systems on big datasets are neural networks and regression models, two well-known
machine learning approaches. The previously described machine learning approaches will be
examined in this part.

 Regression models: Regression models, which are frequently employed in AI forecasting,

seek to create a functional relationship between the target or predicted variable and the
input or predictor variable. By minimizing the difference between the expected and actual
values in the training data, the model gains an understanding of the parameters of this
relationship. Depending on how intricate the relationship being modelled is, regression
models can be either linear or nonlinear. Regression analysis models that exhibit linearity
indicate that there is a linear relationship between one or more independent variables and
one or more dependent variables. On the other hand, more intricate and adaptable
relationships between the variables are possible with nonlinear regression models.
 Neural network: Artificial neurons arranged in interconnected layers make up neural
networks, a class of machine learning models that instruct AI on how to process information
similarly to the human brain. This makes it possible for neural networks to discover intricate
linkages and patterns in the data through a process known as training, improving the
accuracy and dependability of predictions made about the habits of human consumers. The
network modifies its internal parameters during training to reduce the discrepancy between
expected and actual values.

Because AI algorithms are capable of live or incremental learning, they can automatically
adjust and learn from new data to improve projections over time. This makes it possible for
systems to adjust to shifting patterns and dynamics in the data since the AI system can
integrate new data into the current model without having to start from scratch when new
data becomes available. This suggests further that online learning approaches entail
incorporating fresh data into the model's parameters while maintaining the understanding
 gained from earlier datasets. In the end, demand forecasting efforts can produce more
reliable and precise projections thanks to AI algorithms.

AI-powered demand forecasting software

Retalon Retalon is an AI-powered demand forecasting software that uses predictive analytics
to forecast demand with superior accuracy. It accounts for many of the key factors
influencing demand and provides insights into consumer behavior, product trends, and
market demand
Smart Demand Smart Demand Planner is an AI-powered demand forecasting tool that uses machine
Planner learning to predict future demand. It provides a range of features, including demand
forecasting, inventory optimization, and sales analysis.
FutureMargin FutureMargin is an AI-powered demand forecasting platform that uses machine
learning to predict future demand. It provides a range of features, including demand
forecasting, inventory optimization, and sales analysis.
COLIBRI COLIBRI is an AI-powered demand forecasting tool that uses machine learning to
predict future demand. It provides a range of features, including demand forecasting,
inventory optimization, and sales analysis.
amoCRM amoCRM is an AI-powered demand forecasting tool that uses machine learning to
predict future demand. It provides a range of features, including demand forecasting,
inventory optimization, and sales analysis
ClosePlan ClosePlan is an AI-powered demand forecasting tool that uses machine learning to
predict future demand. It provides a range of features, including demand forecasting,
inventory optimization, and sales analysis
Capsule Capsule AI is a sophisticated software that uses machine learning and AI to analyze
data and predict demand in the apparel industry, enabling businesses to make
informed decisions and optimize inventory management.

Comparison of software based on features and brand uses.

 Software Features Brand
 Predictive analytics Levi Strauss & Co.
 Human involvement Nike
 Customizable models Adidas
ClosePlan  User-friendly interface Puma
VF Corporation
H&M
Zara
Gap, Inc.

 Accurate Demand Prediction Nike

 Real-Time Updates Gap
Capsule  Historical Data Analysis Levi’s
 Sales Channel Analysis Under Armour
 Multi-Product Forecasting
Levi Strauss & Co.
 Historical Data Analysis Macy's
 Inventory management Target
 Customer behavior analysis Kohl's
 Real-Time Updates Nike
Amocrm  Integration with other systems such as Adidas
e-commerce platforms and payment Puma
gateways. VF Corporation
H&M
Zara
Gap, Inc.
Urban Outfitters
Abercrombie & Fitch
 Advanced machine learning algorithms Levi Strauss & Co.
for accurate demand forecasting Macy's
 Multi-channel sales data integration for Target
real-time insights Kohl's
COLIBRI  Customizable forecasting models to suit Under Armour
your specific needs. Tommy Hilfiger
 Cloud-based platform for easy access Calvin Klein
and collaboration Coach
 Automated reporting and notifications
for timely decision-making
 User-friendly interface
 Advanced Forecasting Algorithms H&M
 Inventory Management Kelly Devine
  Sales Aid Gap
 Multi-Channel Demand Forecasting Puma
Smart Demand  Seasonality and Trends Alexander Wang
Planner Forever 21
Maria Bartiromo
Zara
Levi's
Nautica

 Analyze historical sales data and H&M

identify patterns to predict future Adidas
demand accurately. Zara
 Real-time data processing Abercrombie & Fitch
Future Margin  Multiple sources of data integration American Eagle Outfitters
 Predictive analytics Calvin Klein
 User-friendly interface Coach
GAP Inc.
Levi Strauss & Co.
Lululemon Athletica
Retalon  Analyzing historical data and market Levi's
trends. Gap
 Real-time updates Forever 21
 Inventory management H&M
 Customizable models Nike
 user-friendly interface Target
 Multiple data sources
 Flexibility –can adapt to changing
market conditions
Software Brand 1 Brand 2

ClosePlan Nike Puma

 Nike uses ClosePlan AI software for  The software is designed to
demand forecasting to predict customer analyze historical sales data and
behavior and optimize its supply chain market trends, alongside other
operations. relevant factors, to create
 The software utilizes AI algorithms to accurate demand forecasts.
analyze historical data, live data  This enables PUMA to better
streams, and other factors to provide anticipate customer demand,
real-time insights into consumer optimize its supply chain, and
demand. minimize inventory levels.
 ClosePlan helps Nike identify potential  PUMA's use of ClosePlan AI
competitive threats and opportunities, software enables it to gain a
enabling it to make data-driven competitive edge through more
decisions and stay ahead of the accurate demand forecasting.
competition.

CAPSULE Gap Under Armour

 GAP integrates CAPSULE data with its  Under Armour uses Capsule AI
internal systems to make more accurate software for demand forecasting
and efficient demand forecasts, which in the apparel industry to analyze
helps the company manage inventory historical sales data and identify
levels and reduce waste. patterns to predict future
 CAPSULE's machine learning algorithms demand.
analyze various factors such as past sales  The software also considers
data, consumer behavior, and market factors affecting demand such as
trends to provide real-time demand weather, events, and marketing
forecasting. activities.
 GAP makes informed decisions on  Under Armour optimize its
product development, manufacturing, production and inventory
and distribution. Overall, GAP leverages planning, ensuring that it has the
CAPSULE's AI capabilities to revolutionize right products in the right
the demand forecasting process in the quantities in the right locations
apparel industry. at the right time.

Amocrm H&M Adidas

 To improve its inventory management  Adidas used amocrm to analyze
and optimize its supply chain. historical sales data, market
 The software leverages machine learning trends, consumer behavior, and
algorithms and historical sales data to other factors to provide accurate
predict future demand patterns, allowing forecasts that help businesses to
H&M to better plan production and optimize their supply chain,
reduce waste. reduce inventory costs, and
 The result is a more efficient and improve their
sustainable business model that benefits overall performance.
both the company and the environment.

Colibri Levi's Calvin Klein

Feature ClosePlan Capsule Amocrm COLIBRI Smart Demand Future Retalon
Planner Margin
Predictive YES NO NO NO NO YES YES
Analytics

Human YES YES YES YES NO YES YES

Involvement

Customizable YES NO NO YES NO YES YES

Models

User-Friendly YES YES YES YES YES YES YES

Interface

Real-Time YES YES YES YES YES YES

Updates

Inventory YES NO YES YES YES NO YES

Management

Historical YES YES YES YES YES YES YES

Data Analysis

Sales YES YES YES NO YES NO NO

Channel
Analysis
Multi- YES NO YES YES YES YES YES
Product
Forecasting

Advanced YES NO YES YES YES YES YES

Forecasting
Algorithms

Sales Aid YES NO YES YES YES NO YES

Seasonality YES NO YES YES YES YES YES

& Trends

Unveiling apparel brands harness AI software

Challenges in AI-Driven Demand Forecasting

 Data availability: AI algorithms require large amounts of data to train on. In the apparel
industry, data can be difficult to obtain and may not be in a format that is easily usable by AI
algorithms.
 Model complexity: AI algorithms can be complex and difficult to understand. This can make
it difficult to troubleshoot problems and ensure that the models are performing as
expected.
 Bias: AI algorithms can be biased, which can lead to inaccurate forecasts. It is important to
carefully select the data that is used to train the models and to monitor the models for bias
on an ongoing basis.

Benefits of AI-Driven Demand Forecasting

 Improved Accuracy: - AI-driven demand forecasting models have been shown to outperform
traditional methods in terms of accuracy [5]. This leads to reduced stockouts, overstocking,
and associated costs.

 Enhanced Inventory Management: - By accurately predicting demand, businesses can

optimize their inventory levels, reducing carrying costs and minimizing the risk of unsold
products.

 Faster Decision-Making: - AI models can quickly process vast amounts of data and provide
real-time insights [5]. This enables businesses to make timely decisions and respond to
market changes rapidly.

 Future of AI-Driven Demand Forecasting in Apparel: - The adoption of AI tools for demand
forecasting in the apparel industry is likely to continue its rapid growth.

Key future trends include.

AI-Powered Supply Chains AI will play a more significant role in optimizing the entire supply
chain, from design and manufacturing to warehousing and distribution, ensuring products are
delivered efficiently and sustainably [5].

 Personalization: - AI will enable apparel companies to offer highly personalized experiences

to customers, tailoring products and recommendations to individual preferences.
 Sustainability: - AI can help reduce waste by predicting demand more accurately and
optimizing production to minimize overproduction.
Reference

(1) Trade cloud. (2021). Demand Forecasting using Artificial Intelligence. [online] Available at:
https://www.tradecloud1.com/en/ai-case-study-1-demand-forecasting-using-artificial-
intelligence/.
(2) Plex Demand Caster. (n.d.). Demand Forecasting: An Industry Guide. [online] Available at:
https://www.demandcaster.com/demand-forecasting-an-industry-guide/#:~:text=Traditional
%20demand%20forecasting%20used%20to [Accessed 24 Oct. 2023].
(3) Artificial Intelligence Demand Forecasting Techniques in Supply Chain Management: A
Systematic Literature Review. (n.d.). Available at:
https://ieomsociety.org/proceedings/2022istanbul/97.pdf [Accessed 7 Sep. 2023].
(4) Google.com. (2021). Available at: https://www.google.com/url?
sa=t&source=web&rct=j&opi=89978449&url=https://mu.ac.in/wp-content/uploads/
2021/02/Logistics-and-Supply-Chain-Management-Sunil-Chopra-
1.pdf&ved=2ahUKEwii0N3Yh4-
 CAxW9RmwGHeFUAbkQFnoECBIQAQ&usg=AOvVaw0jpwy0TarqPMZUHy2pONeG [Accessed
24 Oct. 2023].
(5) Limited, I. (n.d.). 5 ways AI can benefit demand forecasting and inventory planning | Infosys
BPM. [online] www.infosysbpm.com. Available at:
https://www.infosysbpm.com/blogs/supply-chain/5
(6) AI Demand Forecasting: Quick Guide on How to Use It | Intuendi
(7) https://www.leafio.ai/blog/top-5-demand-forecasting-solutions-for-retail/
(8) http://retalon.com
(9) http://research.aimultiple.com
(10) http://theguardian.com
(11) http://mdpi.com
(12) Example of brand Zara:- https://www.aptean.com/en-US/insights/blog/machine-learning-in-
fashion-industry