www.jaguar-aps.

com

Demand-Driven Forecasting

Charles.Novak@Jaguar-APS.com

®
Copyright: Jaguar-APS
 Who we are and what we do….

Some of our clients….

®
Copyright: Jaguar-APS 2
 Introduction
Who needs forecasting?

www.jaguar-aps.com Copyright: Jaguar-APS

Forecasting Where?
 Time series forecasting is applicable to almost all
organizations that work with quantifiable data.
 Retail stores forecast sales
 Energy companies forecast load and reserves, production,
demand and prices
 Educational institutions forecast enrollment
 Passenger transport companies forecast future travel
 Banks forecast new home purchases
 Service companies forecast staffing needs
 Hospitals forecast surgeries
 FMCG forecast demand for their products
 Other …

®
Copyright: Jaguar-APS 4
Introduction: Recent Developments
Demand forecasting drives real value within the supply
chain.

Demand-driven forecasting has become a discipline that

senses, shapes and responds to the real demand.

Predictive analytics used to:

• Uncover patterns in consumer behavior.

• Measure effectiveness of marketing investment.
• Optimize financial performance.
• Shape and proactively drive demand using what-if simulations.
• Sense demand signals and shape the future demand supported
by data mining technologies.

®
Copyright: Jaguar-APS 5
Predictive Analytics and Supply Chains
• Cost perceived too high • Strong potential to transform the
• Focus on investments that provide way in which supply chain
immediate and tangible results. managers lead and supply chains
• Too complex to connect the data operate.
nodes across extended supply • Senior management’s need to grow
chain. the business profitably.
• Big data is a distraction at the • Multi-echelon supply chains require
moment. quick and correct signals to
• Skillsets in supply chain and IT are operate effectively.
limited. • “Heads-up” to help sense, analyze,
• Disconnect between the need and and better respond to market
silo managed IT and other changes.
departments. • Data equals information and
• Demand planning not a Core information equals profit.
Competence. • Pressures to synchronize demand
and supply to understand why
consumers buy products.

Why Not Why Yes

®
Copyright: Jaguar-APS 6
 Data Mining

www.jaguar-aps.com Copyright: Jaguar-APS

Data Mining
 Descriptive analysis  Predictive analysis
 Association - looking for  Forecasting - discovering
patterns where one event is patterns in data that can lead to
connected to another event. reasonable predictions about the
future (This area of data mining is
 Sequence or path analysis - known as predictive analytics.)
looking for patterns where one
event leads to another later
event. Data mining is
 Classification - looking for
new patterns. (May result in a sorting through data
change in the way the data is to identify patterns
organized but that's ok.) and to establish
 Clustering - finding and relationships.
visually documenting groups of
facts not previously known.

®
Copyright: Jaguar-APS 8
Managing by Analytics
 Analytics resolve differences of opinion.
 Initial discussion based on opinions but has to be supported
by numbers.
 Cross-functional involvement to improve alignment.
 Typically, managing by analytics is a MAJOR CHANGE.
 Teams work best when analytics rule discussion.

®
Copyright: Jaguar-APS 9
Demand-Driven Forecasting & Supply Chain

Demand
Orchestration

Demand Demand
Translating Shifting

Demand
Shaping

Demand
Sensing

®
Copyright: Jaguar-APS 10
Monthly Demand-Driven S&OP Process

Decisions
Authorized Game
Recos and Plan
Agenda for Exec.
S&OP
Capacity
Constraints Step 5
Step 4 Exec.
Management
Pre-S&OP S&OP
Forecast
Step 3 Meeting Meeting
Supply
Statistical Planning Demand
Forecasts Step2 Phase
Demand Orchestration
Planning Demand Shifting
End of
Step1 Phase
Month Demand Shaping
Run
Forecast Demand Sensing
Reports Demand Shaping Source: T.F. Wallace

®
Copyright: Jaguar-APS 11
 Source: Charles Chase, SAS

Traditional Demand-Driven

Static analytical methods based on Shift from trend and seasonality to

trend and seasonality. dynamic demand signals

Aggregate level (SKU national, SKU Focus on demand-driven framework of

market, …) shaping and sensing, and orchestrating
demand across products, geographies,
Manual overrides to history and/or channels, and customers.
future – judgemental approach.
Integrated, focused, analytic-driven
Standalone product generation process supported by predictive
strategies and rigid monthly process. analytics, market intelligence, more
sophisticated technologies.
Does not capture changing market
dynamics. ‘Real-time’ forecasting based on
market volatility and dynamics –
sensing demand signals weekly and
managing demand orchestration daily
What is Demand-Driven Forecasting? for rapidly changing markets.
®
Copyright: Jaguar-APS 12
Inside-out (Push) strategy issues
“If replenishment takes care of inventory problems,
what caused the inventory problems in the first place?”

 Downstream demand
accumulated ad presented
Store
as aggregate total.
Vendor DC Store  Delay in the initial demand
from original customer.
Store
 Service level need is an
Information
average.
 Upstream supply expected to
Inventory be at 100% service level.

®
Copyright: Jaguar-APS 13
Supply Chain Trends… Evolution from Push to Pull
Push / Inside-Out Pull / Outside-In
 SELL what you MAKE
 MAKE what you can SELL
 Is the demand the result of
consumption at a store level?  From INSIDE-OUT to OUTSIDE-
 Is the demand upstream the result IN
of the downstream consumer
demand or the downstream  It is about owning the
warehouse demand?
 How long it is before a forecast
entire supply chain,
becomes true demand? including the channel, and
 Does the demand signal become managing products from
true just because you believe it to
be better? their manufacture to their
 Do you have enough time to react end use.
to the true demand when it finally
becomes known?

®
Copyright: Jaguar-APS 14
 Data Analysis
Data Patterns – Attributes and Variables

www.jaguar-aps.com Copyright: Jaguar-APS

Application of SPC to Forecasting
 SPC (Statistical Process Control) focuses on Variables and
Attributes in the dataset.
 Attributes data = specific values that we DO EXPECT in our
data
 Level, Trend, Seasonality, Moving Holidays, Promotions, Competitive
Activity, etc.
 Variable data = any value that we DO NOT EXPECT / WANT
in our data
 Out-of-Stock, Trend Intervention, Irregularly Scheduled Promotion,
Competitive Activity, Economics, etc.
 Normal Distribution - Control charts

®
Copyright: Jaguar-APS 16
Understanding Your Data
 Line graph is ideal for visualization of time series.

®
Copyright: Jaguar-APS 17
Recognizing Data Attributes

Level

®
Copyright: Jaguar-APS 18
Recognizing Data Attributes

Level Trend

®
Copyright: Jaguar-APS 19
Recognizing Data Attributes

Level Trend

Seasonality

®
Copyright: Jaguar-APS 20
Recognizing Data Attributes & Variables
Total All 36 months Last 24 months Last 12 months

Trend @ 36, 24, & 12 periods

®
Copyright: Jaguar-APS 21
Recognizing Data Variables
Pipeline Fill – Unrealistic Trend

Expected Trend

®
Copyright: Jaguar-APS 22
Recognizing Data Variables
Negative Sales

®
Copyright: Jaguar-APS 23
Recognizing Data Variables
Trend and intervention

®
Copyright: Jaguar-APS 24
Recognizing Data Variables
Missing Data

®
Copyright: Jaguar-APS 25
Recognizing Data Variables
Lumpy / Intermittent Demand

®
Copyright: Jaguar-APS 26
Recognizing Data Variables
High Variability / Short History

Sales Average

®
Copyright: Jaguar-APS 27
Data Patterns Is this product stable or dynamic?

Are there any changes in

data that we need to
build in the forecast?

Are there any issues we

need to correct?

Is this product trended? Is this product seasonal?

®
Copyright: Jaguar-APS 28
 With help of SPC and visualization of the data, the
Are there any changes in questions can be answered with more accuracy than just
Data
data that Patterns
we need to a gut feel.
build in the forecast?
Is this product stable or dynamic?

Are there any issues we

need to correct?

Is this product seasonal?

Is this product trended?
®
Copyright: Jaguar-APS 29
Recognizing Data Variables: Sigma Alert
 Standard deviation of last six months history is significantly
higher/lower than previous 18 months.
 Indicates change in recent history that needs to be
understood and either cleansed or incorporated in forecast
adjustments or forecast model.
SIGMA ALERT

250,000.00

200,000.00

150,000.00

100,000.00

50,000.00

-
P1 P7 P13 P19 P25 P31

Sales Cleansed History

®
Copyright: Jaguar-APS 30
Data Pre-Processing / Cleansing
 Extreme Values
 Unusually large or small compared to other values in series
(Outliers).
 Unusually different values within the series close to mean
(Inliers).
 Automatic versus manual correction of Extreme Values
 Most forecasting systems can do it. Not all can do it well. Use
judgement and domain knowledge to remove or keep it.
 Choice of Time Span
 How many historical periods will be used for the forecast?
 Simple Exponential Smoothing and Averages need 2-3 values
 Double Exponential Smoothing needs 4 values
 Triple Exponential Smoothing needs 24-36 monthly or 154 of weekly
values
®
Copyright: Jaguar-APS 31
SPC Control Chart: Identifying Patterns
Upper Control Limit

Outlier
Variations
Normal
Center Value

Outlier
Lower Control Limit

Trend ???

Center value and UCL/LCL definitions: There is a difference between Mean and Median,
1𝜎 and 3 𝜎, … Rule of thumb – start with Mean 18 Months and 1.5 𝝈
®
Copyright: Jaguar-APS 32
 Data Segmentation

www.jaguar-aps.com Copyright: Jaguar-APS

Segmenting Products to Choose Appropriate
Forecasting Method
 Time series analysis  demand patterns  forecastability.
 Value to the company + forecastability  correct
forecasting method.
Data Patterns

Trend, Seasonality,
Data
Cycle, Randomness Forecastability
Data Segmenation

+ Forecast Methods
Value Value to the
Company
High / Low

+
Marketing

New, Harvest,
Growth, Niche

® Source: Charles Chase, SAS

Copyright: Jaguar-APS 34
 Source:

Data Segmentation - Forecastability

High Volume / High Variability
High

• Seasonal Items
High Volume / Low Variability
• Promotions
• High Runners
• Short Lifecycle
• Commodities
• Determine if variability is
• Predictable
predictable (Seasonal)
• Disruption Spikes
Volume

Low Volume / Low Variability Low Volume / High Variability

• Consistent low runners • Specialty Items
• Longer Lifecycles • Custom Orders
• Predictable • Can Have High Margins
• Disruptions can have big impact • Consistent Disruptions
Low

(often tied to lead time) • Often Tied to Other SKUs

Low Demand Variability High

®
Copyright: Jaguar-APS 35
 Data Segmentation - Forecastability
LOW VARIABILITY, LOW
VOLUME, 287
LOW VOLUME

HIGH VARIABILITY, LOW

VOLUME, 393
LOW VARIABILITY, HIGH
VOLUME, 38
HIGH VOLUME

HIGH VARIABILITY, HIGH

VOLUME, 16

LOW VARIABILITY HIGH VARIABILITY

®
Copyright: Jaguar-APS 36
Four Quadrants Based on Portfolio Management
New Products Growth Brands
High Value High Value
Low Product Line Extensions: High Priority Products: High
Forecastability (Evolutionary New Products). · Strong Trend Forecastability
Some ‘Like’ history available. · Seasonal Fluctuations
Short Life Cycle Products: · Possible Cycles
Many ‘Like’ products available. · Sales Promotions
New Products: · National Marketing Events
(Revolutionary New Products) · Advertising Driven
Company Value

No ‘Like’ history available. · Highly Competitive

Niche Brands Harvest Brands

Low Priority Regional Specialty

Low Priority Products:
Products:
· Strong Trend
· Some Trend
Low Value · Highly Seasonal Low Value
· Seasonal Fluctuations
Low · Possibly Cycles High
Forecastability · Irregular Demand Forecastability
· Minor Sales Promotions
· Local Targeted Marketing
Events

Forecastability
® Source: Charles Chase, SAS
Copyright: Jaguar-APS 37
Statistical Methods Selection Based on Segmentation
and Portfolio Management
Judgmental Causal Modeling
New Products Growth Brands
High Value High Value
Low High
ARIMAX
Forecastability Forecastability
‘Juries’ of Executive Opinion ARIMA with Interventions and
Delphi Committees Regressors
Sales Force Composites Simple Regression
Independent Judgment Multiple Regression
Company Value

Multiple Methods Time Series

ARIMA Box-Jenkins
Combined average:
Winters 2 / 3 Parameter
Niche Brands Judgment, Time Series, Causal Harvest Brands
Decomposition
Low Value Combined Weighted: Low Value
Low Simple Moving Average High
Judgment, Time Series, Causal
Forecastability Holt’s Double Exponential Forecastability
Croston’s Intermittent Demand
Smoothing

Forecastability

® Source: Charles Chase, SAS

Copyright: Jaguar-APS 38
 Forecasting Outside-in
Linking Market Data to Shipments – Simplified Example

www.jaguar-aps.com Copyright: Jaguar-APS

MTCA – Multi-Tiered Causal Analysis
“Integrating consumer demand into the demand forecasting
process to improve shipment (supply) forecasts has become a
high priority in the FMCG/CPG industry as well as in many other
industries over the past several years.” Charles Chase, SAS

 Past constraints are becoming non-issue today:

 Data collection and storage
 Computing power available
 Data synchronization capabilities
 Analytical expertise
 MTCA, a process of nesting causal models together using
data and analytics, considers marketing and replenishment
strategies jointly, rather than creating two separate
forecasts.
® Source: Charles Chase, SAS
Copyright: Jaguar-APS 40
Growth brand – traditional forecasting approach

®
Copyright: Jaguar-APS 41
MTCA – Multi-Tiered Causal Analysis

Market data from ACNielsen

Factory shipments

®
Copyright: Jaguar-APS 42
1. Data Analysis

Comparisson of category consumption and forecast with all outlet consumption for
brand. Strong correlation between the two variables is confirming origninal assumptions
of category influencing brand consumption.

®
Copyright: Jaguar-APS 43
 2. Development of Consumption Forecast
Regression 1: Last 4 Periods Out of Sample
b 0.670646293 114991.935 a Running two regressions: first to validate the
SE X 0.01659508 43476.00362 SE Y
R2 0.989098563 57723.81787 SE model, second to use the model to generate
F 1633.158501
SS Reg 5.44175E+12 59976704694
18 df Residual
SS Residual forecast for the brand‘s consumption.
t 40.41235579

Out of Sample
Forecast Bias % Bias Abs Error APE
4,247,556 329,111 7% 329111.089 7%
1,997,206 (131,092) -7% 131091.571 7%
1,772,187 (44,406) -3% 44405.9146 3%
1,724,543 (11,602) -1% 11601.8607 1%
ME -1% MAPE 4%

Regression 2: All Data In Sample

b 0.716448119 -987.7474012 a
SE X 0.015630716 44552.36932 SE Y
R2 0.989636968 78137.09794 SE
F 2100.930716 22 df Residual
SS Reg 1.2827E+13 1.34319E+11 SS Residual
t 45.83591076

®
Copyright: Jaguar-APS 44
3. Development of Shipment Forecast Based on
Consumption Forecast Detection of lag/lead relationship of factory
shipments and consumption.

Shipment forecast based on consumption.

®
Copyright: Jaguar-APS 45
4. Linking Consumption Forecast to Supply Chain and
Internal Marketing/Sales Programs
Adding supply events and TV
advertising (dummy variables) plus
marketing promotions (past history
and forecast) as final variables to
the consumption based factory
shipment forecast.

Final forecast (green) based

on consumption, supply chain
constraints, marketing and
sales activity.

®
Copyright: Jaguar-APS 46
Comparison of traditional forecasting approach (blue)
versus MTCA (red).

What would it mean to supply chain if only statistical shipment forecast was used?

®
Copyright: Jaguar-APS 47
Why Haven’t Companies Embraced the Concept of
Demand-Driven?

Traditional view of
Incentives supply chain
excellence

Leadership Focus: Inside out, not

outside in

Vertical rewards Focus on transactions

versus horizontal not relationships
processes

® Source: Charles Chase, SAS

Copyright: Jaguar-APS 48
Why Haven’t Companies Embraced the Concept of
Demand-Driven?
 Incentives:
 As long as sales is incented only for volume sold and marketing only for market share, companies will never
become demand driven. To make the transition to demand-driven, companies must focus on profitable sales
growth through the channel.
 Traditional view of supply chain excellence.
 For demand-driven initiatives to succeed, they must extend from the customer's customer to supplier’s supplier.
Customer and supplier initiatives usually are managed in separate initiatives largely driven by cost.
 Leadership.
 The concepts of demand latency, demand sensing, demand shaping, demand translation, and demand
orchestration are not widely understood. As a result, they are not included in the definition of corporate strategy.
 Focus: Inside out, not outside in.
 Process focus is (today) from the inside of the organization out, as opposed to from the outside (market driven)
back. In demand-driven processes, the design of the processes if from the market back, based on sensing and
shaping demand.
 Vertical rewards versus horizontal processes.
 In supply-based organizations, the supply chain is incented based on cost reduction, procurement is incented
based on the lowest purchased cost, distribution/logistics is rewarded fro on-time shipments with the lowest costs,
sales is rewarded for sell-in volume into the channel, and marketing is rewarded for market share. These
incentives cannot be aligned to maximize true value.
 Focus on transactions not relationships.
 Today, the connecting processes of the enterprise – selling and purchasing – are focused on transactional
efficiency. As a result, the greater value that can happen through relationships – acceleration of time to market
through innovation, breakthrough thinking in sustainability, and sharing of demand data – never materializes.

® Source: Charles Chase, SAS

Copyright: Jaguar-APS 49
Recommendations
 Understand Demand
 By better understanding demand, companies can plan production
capacity and inventory level in a more accurate fashion, minimizing
the risk of lost sales opportunities.
 Collaboration and Integration
 The ability to share information between departments within the
business is essential to improving supply chain.
 Without internal communication processes in place (Demand Planning,
S&OP), the company as a whole cannot effectively collaborate with
the outside entities, whether they are supplier or customers.
 Supply Chain Management
 Increased visibility into supply decisions and constraints by providing
input in the demand shaping and shifting activity will help ensure the
product is available at the right place at the right time.

®
Copyright: Jaguar-APS 50
 Thank You 
Charles Novak
charles.novak@jaguar-aps.com
Pavel Černý
pavel.cerny@jaguar-aps.eu
www.jaguar-aps.com

Our Solutions http://www.jaguar-aps.com/services/index.html

• Training Programs
• Opportunity Assessment
• Business Transformation – S&OP/IBP
• Forecast Software Selection, Tuning and Implementation
• On-Demand

®
Copyright: Jaguar-APS