Improving Forecasting for the Supply Chain

1
Improving Forecasting for the Supply Chain

• Robert Fildes, Centre for Forecasting, Lancaster
  University Management School
• President, International Institute of Forecasters

2
Hierarchical Forecasting in the Supply Chain
Macro variables
e.g. growth
Consumer demand
e.g POTS
Business
demand
Aggregate
Competitors'
Company demand
demand
Distribution Centres Retailers
Product
Product
Product
Class A
Class B
Class C
1
2
3
3
Improving data flows
Forecasting for Production Inventory
Market factors
Product Classes
Class A
Class B
Total
AA
AB
AC
AX
BJ
. . . . .
. . . .
Products
Labour
Machines
Parts
Raw Materials
4
THE RESULTS of POOR FORECASTING

• SHORT TERM
• Stock/ service level
• MARKET PLANNING
• Capacity problems
• Pricing sales force management
• inefficient financial management
• LONG TERM TECHNOLIGICAL/SOCIAL CHANGE
• Bankruptcy

5
EVALUATING THE FORECASTING ACTIVITY

• Decision Effectiveness

• Accuracy
• bias
• variance

• Cost
• Speed

• Motivational Implications
• Sales Force Remuneration
• Feedback/ Self-fulfilling Prophecy
• Goal Signalling

Forecasts are Frequently Politically Modified
6
Forecasting Requirements down the Supply Chain
Additional Information from the Supply Chain
Customers

• What must be forecast
• Customer demand
• Retailer orders
• Manufacturer Orders

Sales

Retailer
Shipments
Orders
Manufacturer
Parts Materials Suppliers
7
Information affecting the supply chain
The basic model Ordersf(past orders)
judgemental estimates of promotions The full
model Ordersf(past orders, Sales, forecast
sales, promotions, Events)
8
Linking with the Retailer

• Sharing data
• Sharing plans
• Sharing forecasts

The benefits?
9
The Benefits of Additional Information?

• Sources of Additional Information
• from cross-correlations (steal)
• from common factors, i.e seasonality
• (Bunn Vassilopoulis, IJF, 1993)
• from orders
• from external factors
• from the marketing mix
• from managerial information

10
The Value of Forecasting

• Mixed evidence
• value depends on absolute error level
• depends on production system
• depends on service level, cost trade-off
• Achievable accuracy improvements
• 30 possible

Service - inventory investment tradeoff curves
Service
Inventory Investment
11
Typical Approach to Forecasting for the Supply
Chain

• Data is often unstable

• Statistical forecast obtained
• usually exponential smoothing type approach
• rolling forecasts are used
• Managerial judgement then used to adjust the
  forecast

12
BENCHMARK FORECAST ERRORS (MAPE)
But these figures are too low ? 50 - 100 for
low frequency, 15-30 1 month for product demand,
retail
The Laws of Forecasting Forecast as short a
period ahead as possible Forecast at the
highest level of aggregation possible
13
IMPROVING the ORGANIZATION of FORECASTING
Activity
Respondents Scoring Important

• 83
• 70
• 66
• 61
• 35

• Developing consistent data
• Increased software support
• Improved techniques
• Improved data bases
• Improved communication with users

14
The Supply Chain Forecasting Process
System Variables
Data
Previous Forecast Error
Method based forecast
Judgemental adjustment
Compare
Judgemental forecast
Final Forecast
Additional market information forecasts -
by category total
How do organisations integrate different
information sources?
15
How to design and manage the forecasting
processto deal with market complexity
Issue

• Staff
• Motivation
• Training
• Information
• Data base Key variables collected regularly
• Systems
• The design and use of FSS
• Organisational aspects
• Value of good forecasting recognised
• Information flows facilitated through integration
• Responsibility of accurate and unbiased forecasts
  transparent

• Moon Mentzer (IJF, 03)
• functional integration
• Approach
• systems
• Performance measurement

16
Issue Staff

• Technical staff
• There arent any!
• Forecasters
• No training
• Limited aspect of job for most
• Bias (Sanders Manrodt, Stewart)
• Not appraised
• Users
• Ambivalent about the possibility of achievable
  improvements
• Political nature of forecasting

Certification and training?
17
Issue Information Organisational Aspects

• No responsibility for collection of forecast
  oriented information
• Pools of analysis.
• Information collected in different parts of the
  organisation is not transferred.
• No clear organisational responsibility for the
  forecasting function
• Location?
• no learning, no improvement

18
Motivations Affecting Forecast Accuracy- the
Agency Problem

• Use of high forecasts to support funding requests
• Use of low forecasts to increase performance
  related pay
• Desire (by operations or development) to hide or
  ignore product limitations corrupts data
• Use of extreme forecast to achieve greater
  recognition
• Benefits from being extreme
• Financial prudence
• Ideology

19
Forecasting Support Systems (FSS)
Issue how to design and manage the forecasting
process to deal with market complexity

• Systems designed to support forecasting by
  providing
• statistical methods,
• facilities for formulating informed management
  judgments,
• facilities for the integration of statistical
  forecasts with management judgment.And
• Possibly an extended information set, e.g. prices

A Type of Decision Support System
20
Complementary nature of statistical forecasts
management judgment
Dealing with the complexity Organisationally
based Forecasting combines statistical analysis
with managerial judgement

• humans are adaptable and can take into account
  one-off events, but they are inconsistent and
  suffer from cognitive biases
• statistical methods are rigid, but consistent,
  and can take into account large volumes of
  information

21
Combining Statistical, Customer Managerial
Forecasts
Forecasting Support System

• Customer forecast used for first two periods
• Customer forecast compared with actual for
  accuracy
• Statistical and MI are compared with actual for
  accuracy
• Separate accuracies are compared and used to
  improve process
• Communication process is very important

The Effective System relies on Combining
Different Information Sources
22
The Stories
Issue Forecasting systems

• No statistical basis of models
• No ability to explore alternative models
• Limited tailoring to user requirements
• System used in default mode
• No corporate technical knowledge
• Poor measures of performance
• no benchmarks

• User interventions unstructured
• No monitoring of effectiveness of user
  interventions
• No history of interventions

23
The Consequences of Poor Forecasts

• Too much stockor
• Unnecessarily Poor Service

Service - inventory investment tradeoff curves
The wrong product in the wrong place at the wrong
time
24
Examining Sales Forecasting Practiceto Improve
Supply Chain Forecasting - Research Programme

• Product hierarchy data base
• Organisational Requirements for operations
• Information flows availability
• Forecasting Methods
• Current Accuracy Levels
• Users
• User interventions their value
• Motivation
• Design involvement
• System issues

Can we design better processes and systems to
Improve Accuracy and Effectiveness?
25
Examining Sales Forecasting Practice

• Frequency of forecasting
• Number of products forecast
• Forecast horizon
• Information available
• a)       sales history
• b)       price promotions
• c)       past forecasts
• d)       Account managers forecasts
• Lowest Level of Aggregation
• Organisational Importance of forecasting
• Uses/ users of Forecasts

• Involvement in system choice
• Involved in system design
• Range of methods
• Choice of method
• Parameterisation
• Nature of Judgemental Intervention
• Reliance on system forecasts
• Documentation of intervention
• Accuracy considerations
• Perceived System weaknesses

26
Ideal Use of Support System

• delegating to the system routine computations
  and resolutions of interaction s too complex for
  the manager to perform
• While
• leaving the judgements that the algorithm could
  neither make, nor recognize were needed, to the
  human
• Keen Scott Morton, 1978

The FSSs role is to effectively integrate the
statistical methods with managerial judgement
27
Consequences of non-ideal use of FSS

• Judges read noise as systematic
• Statistical forecasts distorted by transitory
  special events
• Double counting of some effects
• Basis of forecast is unclear and cannot be easily
  communicated
• Wasted managerial effort
• Managers may have an effort budget

28
Objectives of this research programme

• To understand the existing and potential design
  features which are conducive to the adoption,
  acceptance and effective use of FSS by
  forecasters.
• To investigate the use of forecasting support
  systems (FSS) in companies to establish the role
  they play in forecasting processes and the extent
  to which their role can be improved
• to improve the effectiveness (and accuracy) when
  FSS are used to combine the strengths of
  statistical methods with managerial judgment
• In addition, a methodological contribution is
•         To compare FSS usage in an
  organisational setting with experimentally based
  evidence.

29
Supply Chain Forecasting the Key Issues

• Inclusive data base
• Collaboration
• Recruitment and Training of Staff
• A Certificate in Forecasting Practice?
• Organisational Issues
• Motivation information sharing
• Organisational improvement through learning
• The Forecasting Support System

30
Conclusions

• Value of improved forecasting in supply chain
• substantial in situations with high noise
• Major Improvements possible through design of FSS
• Managerial judgment problematic
• But increases user acceptability
• accuracy?
• special effects?
• Organisational Responsibilities
• User motivation
• Performance measurement
• Training (selection)

Despite the advances in statistical forecasting
techniques and software, performance has
improved little if at all (Moon et al, 2003)
31
What goes wrong?

• In a phrase
• the organisation does not care enough!
• - its culture does not support learning

32
Lancaster Centre for Forecasting

• Research
• sponsorship of projects
• Contract Research
• market analysis and forecasting
• price elasticity estimation
• software appraisal/ development
• call centre
• pricing
• company based Organisational Forecasting Audit
• MSc Projects
• 4mths, agreed brief, expenses
• Seminar Programme Forecasting Practitioner
  Network
• Research Programme in Using Software Effectively

Reference Armstrong, J.S. (ed.) The Principles
of Forecasting, Kluwer, 2001.
Director Professor Robert Fildes, Lancaster
Centre for Forecasting, Lancaster University, LA1
4YX Tel (44) (0) 1524 - 593879 email
R.Fildes_at_Lancaster.ac.uk
33
(No Transcript)
34
System Variables
Data
Previous Forecast Error
Method based forecast
Judgemental adjustment
Additional forecasts - by category total
Compare
Judgemental forecast
Final Forecast
Figure 1 The Forecasting Support System