FAR001
It All Starts With a Forecast: An Overview of JDA Demand Paula Natoli
Demand Management The Problem The expansion of product assortments, the shortening shortenin g of produc productt lifecycles, the proliferation of promo promotiona tionall offerings, offerings, and the need for dynamic pricing strategies leads to an increased increased complexi complexity ty of being able to accurately predict customer demand. demand.
The Solution Understand and predict customer demand by product, location, and time in order to maximize maximi ze sales effectiveness while minimizing inventory expense.
Demand Management The Problem The expansion of product assortments, the shortening shortenin g of produc productt lifecycles, the proliferation of promo promotiona tionall offerings, offerings, and the need for dynamic pricing strategies leads to an increased increased complexi complexity ty of being able to accurately predict customer demand. demand.
The Solution Understand and predict customer demand by product, location, and time in order to maximize maximi ze sales effectiveness while minimizing inventory expense.
Demand Management Solution Suite Components
Business Objectives
Demand
Demand Manager
Demand Decomposition
Demand Planner
Demand Classification
Profiler
Dynamic Demand Response
Seasonal Profiling
Market Manager
Accurate, efficient time time phased forecasting KPI Driven Planning
Key Capabilities
This document contains forward –looking –looking statements based on current expectations, forecasts and assumptions
Consolidated view of customer demand
Automated history classification, classification, algorithm selection and parameter setting
Comprehensive set of statistical & configurable business rules for forecasting
Hierarchical, multi channel forecasting
Robust, Flexible Demand Analysis
Differentiate between seasonal and promotional lift in sales history
Identification of common seasonal profiles
Integrated Promotional planning & analysis
Sense & Respond Order Pattern Recognition
Support for consensus based planning
Demand Management: One View of Synchronized Demand S&OP
Historical Sales Information
Replenishment & Allocation Space & Category Management
Sales History Lost Sales
Demand Management
Consensus Forecast
Workforce Management
Pricing / Promo Effects
Trade Promotions Management
Causal Factors
Trend
Merchandise Planning
Pricing & Promotions Management & Optimization
Seasonality Internal / External Collaboration
Transportation Management
Demand Key Features and Highlights Comprehensive Algorithm Library • Eight statistical algorithms to properly handle differing demand patterns • Demand Classification classifies historical patterns, assigns appropriate algorithm(s) and tunes parameters.
Multi-Channel, Flexible Hierarchical Demand Management • Flexible hierarchy configuration extending product, channel, and location hierarchies • Supports bottoms-up, tops-down or middle-out aggregation & reconciliation • Attach Rate Forecasting: Independent + Dependent Demand Forecasting
Lifecycle Management & New Product Introduction • Comprehensive new product/store introduction functionality • Creation and management of lifecycle curves and launch profiles for re -use • Attribute based short lifecycle forecasting
Tight Integration with Complimentary Solutions • Shared data model and UI navigation with Fulfillment, Pricing/Promotions, Master Planning • Common foundation with extended suite: Collaborate, S&OP
Where do I start? • T h e w a y y o u r e p l en i s h i s n o t n e c e s s a r i ly t h e same way you s ho uld forecast.
• Define what entity you will forecast. – SKU/Store – SKU/Warehouse – SKU/Region – Category/Region – SKU/Customer/Warehouse – Etc…
Demand Forecasting Unit (DFU) At its simplest, a DFU represents an item selling in a market
Product
Market
Demand Forecasting Unit Markets are often further delineated… Product
Market
Channel
Region
creating the complete DFU: Product
Channel
Region
Demand Forecasting Unit DFU Examples
Product
Channel
Region
20 oz. Cola
Convenience Stores
New York
Consumer Goods 12 oz. Cola
Superstores
South East
MassMerch Retail Laundry Detergent
Retail
Store 123 Power Drills
Internet
West Region
Demand Forecasting Unit + Model Product
DFU Examples
Channel
Consumer Goods
12 oz. Cola
Superstores
South East
Convenience Stores
20 oz. Cola
FourierOrder LewandowskiShipments
New York
MassMerch Retail Laundry Detergent
Retail
Power Drills
Store 123
Internet
MLRPOS
West Region
MLRShipments
Region
Model
Model represents a combination of history and statistical algorithm
Multi-level Forecasting Process • Organizes your DFUs into a manageable structure • Synchronizes data among multiple levels through aggregation and reconciliation models • Facilitates the inclusion of business knowledge into the forecasting process
Multi-level Forecasting
Channel Product Category Class Sub Class SKU
Furniture
Storage
Chairs
High Back
10001
Filing
Low Back
10002
Region
Multi-level Forecasting
Region
Product Channel All
All Channels Market/Distribution Channels
Retail
Catalog
Internet
Multi-level Forecasting
Product
Channel Region All
All Warehouses Ship From Warehouse (e.g. North America) Minor Region (e.g. SouthEast United States)
DC1
DC2
NY
DC3
Bos
Defining the DFU Structure A DFU can be any connection across these three dimensions. Product
Channel
Department
Channel
Region
Country
Category
Store Cluster
Class
Store
Item
Each defined DFU level can generate its own statistical forecast based upon aggregated sales history.
Aggregation Model Most aggregate level
d n a m e D A D J
e r u t c n u o r i t t S a l g e e v r e g L g i A t l u M
Granular level
• Aggregate History through the nodes in a hierarchy • Generate independent forecast at each level in hierarchy • Hierarchies can be in other dimensions – Channel & Location
Reconciliation Model Top Down Most aggregate level
d n a m e D A D J
e r u t c u e r l i t S c l n e o v c e e L R i t l u M
Granular level
• Enables comparisons of statistically generated forecasts at multiple levels • Allocates time phased forecasts across defined hierarchy levels • Flexible definition of reconciliation models
Evaluate and Prepare
Demand Management Process •
Determine which levels of the product and business hierarchy will be forecasted
•
Determine the historical stream: – – – – – – –
•
Consumer Demand (POS) Store Orders Retail DC to Store shipments Retail orders Adjusted orders Shipments Events
History Cleansing – – – –
Base vs. Non-base De-spiking Lost Sales Pricing/Promotional Effects
History
Calculate Model using forecasting method
Evaluate and Fine Tune the Model
Evaluate Forecast Performance
Manage Exceptions
Manage Additional Information
Close the Period
Draft Forecast
Publish Forecast
• Amount of history •
Unit of measure
r e s h t e o s s o e t c d o n r e p S
Consensus Forecasting
Demand Decomposition Separate Base from Promo History Key Actions
Separate base history from promotional business and seasonal peaks/valleys
Estimate impacts of promotional history
Provide a cleansed historical data stream for which to apply statistical models
Business Impacts
Develop a solid baseline statistical forecast
Properly identify gaps in the future forec ast and plan future events accordingly
Understand the impact that promotions play on your business
Demand Management Process
Evaluate and Prepare History
Calculate Model using forecasting method
• Identify algorithm to use via D em a n d C l as s i f i c a t i o n – Assigned to each forecasting entity (i.e.: item/channel/loc) – Recommend appropriate settings of the model parameters •
•
•
Process to select algorithm is run only 1-4 times a year Process to tune parameters can be run more frequently Output is a baseline statistical forecast
Evaluate and Fine Tune the Model
Evaluate Forecast Performance
Manage Exceptions
Manage Additional Information
Close the Period
Draft Forecast
Publish Forecast r e s h t e o s s o e t c d o n r e p S
Forecasting Methods Regression Based Algorithms • Fourier assumes that business is constant or that it changes at a constant rate •Multiple Linear Regression first develops a profile assuming constant business, then develops the model further through the use of multiple causal variables.
Smoothing Based Algorithms • Lewandowski method assumes that not only is business inconstant, but also that it changes at an inconstant rate. •Croston Used for slow, lumpy items; (i.e.: many zeroes) •Holt Winters Most useful when the seasonal component and the trend component are changing at different paces over time. •Moving Average assumes near term forecast to be similar in nature to recent past • AVS-Graves handles varying intermittent demand patterns with ability to incorporate seasonality
Other • Short Lifecycle utilizes attribute based mapping to extract existing lifecycle curves and apply them to new DFUS
Sales Patterns Differ Across Products Groups and/or Locations! What type of products do you deal with? • continuous vs. intermittent • seasonal vs. non-seasonal • trend vs. constant • stable vs. highly variable A mixture of “all of the above”?
When it comes to statistical algorithms, one size does NOT fit all !
Demand Classification Assignments by Class
Review Percent of DFUs Per Class
• Classify products in terms of their historical demand pattern
Demand Classification Assignments by Algorithm
Review Recommended Algorithms for DFUs
• Automatically assign the recommended algorithm and starting parameters based on history patterns • Reduce planner fine-tuning time
Generate the Statistical Forecast Statistical forecast is generated based upon the sales history at each defined DFU level
1. Map the historical sales data 2. Fit a model to the history (fitted history) 3. Project the model into the future (forecast )
Demand Management Process
Evaluate and Prepare History
Calculate Model using forecasting method
•
Review system suggestions for parameter settings Evaluate and Fine Tune the Model
– Fine-tune suggested parameter settings where appropriate Evaluate Forecast Performance
•
Manage Exceptions
Each algorithm supports a unique set of parameters Manage Additional Information
Close the Period
•
Each DFU can have parameters adjusted independently
Draft Forecast
Publish Forecast
r e s h t e o s s o e t c d o n r e p S
Forecasting Demand Workbench Configurable Panels Graphical Hierarchy Navigation
Configurable Panels
New Product Introduction & Lifecycle Management Launch Manager: Allocate forecasts for new launches following previous similar product launch profiles
New Product Introduction: create new DFUs & generate initial forecasts based upon “like” DFUs.
Lifecycle Manager:
Create, Manage, Extract and Assign Lifecycle profiles to DFUs
Short Lifecycle Forecasting • Automated forecast generation for products with short lifecycles • Bypass need to manually select lifecycles to attach to new products • Works upon attribute matching and Bass curve modeling & fitting
Define & Prioritize DFU Attributes Define the DFU attributes that will most likely represent similar traits for new DFUS.
Build Short lifecycle curves Builds and stores curves for selected DFU groupings.
Assign Short Lifecycle Curves Adjust the forecast based on the history and tuning
Adapt Forecast
Matches new DFUS to stored curves based upon attributes.
Demand Management Process
Evaluate and Prepare History
Calculate Model using forecasting method
• Review modeling exceptions • Tune parameters • Regenerate forecasts
Evaluate and Fine Tune the Model
Evaluate Forecast Performance
Manage Exceptions
Manage Additional Information
Close the Period
Draft Forecast
Publish Forecast
r e s h e t o s s o e t c d o n r e p S
Forecasting Exception Graphs •
Plotting of DFUS graphically to understand and manage by exception
•
Multiple graph types support various exceptions Immediate Visibility to Changes
Demand Management Process
Evaluate and Prepare History
Calculate Model using forecasting method
• Adjust baseline forecasts to account for market insights – Promotional Information
Evaluate and Fine Tune the Model
– Competitive Information – Forecast Overrides – Cannibalization
Evaluate Forecast Performance
Manage Exceptions
– Business Weather Intelligence
•
Supports integration with Promotions & Pricing Management suite
Manage Additional Information
Close the Period
Draft Forecast
•
Incorporate changes from a consensus forecasting business process
Publish Forecast
r e s h e t o s s o e t c d o n r e p S
Consensus Forecasting
Incorporate Event Information: A Building Block Approach • Utilize statistical forecast to establish a baseline of current mean, trend, seasonality • Incorporate promotional lifts • Add marketing intelligence, cross-functional information to align the organization to the game plan
Consensus Adjustment Total Forecast
Promotional Lift Baseline Statistical Forecast
Forecast Overrides Apply incremental overrides to individual forecasts or flexible groupings of forecasts
Incorporate Promo Lift: Promotions Management
Market Activities One View of Demand
Promotion lift incorporated directly into the forecast
Forecast Reconciliation Tops-Down / Bottoms Up •
Reconcile forecasts via tops/down or bottoms/up
•
Supported in both a fixed hierarchy (multiple levels) and a flexible grouping
View Percentage Contribution
Total of selected DFUS
Evaluate and Prepare History
Demand Management Process Calculate Model using forecasting method
• Transfer forecasts to other required business processes & applications
Evaluate and Fine Tune the Model
– Fulfillment – Enterprise Planning
Evaluate Forecast Performance
Manage Exceptions
– Space & Category Management – Workforce Management Manage Additional Information
Close the Period
• Enables all other processes to work from one view of demand
Draft Forecast
Publish Forecast
r e s h e t o s s o e t c d o n r e p S
Evaluate and Prepare History
Demand Management Process
Calculate Model using forecasting method
• Bring in history from posted period
Evaluate and Fine Tune the Model
• Regenerate updated forecast based upon added history
Evaluate Forecast Performance
Manage Exceptions
Manage Additional Information
Close the Period
• Review forecast accuracy metrics from completed period
Draft Forecast
Publish Forecast
r e s h e t o s s o e t c d o n r e p S
Forecast Performance Average Forecast Error
Plan Production
Purchase Materials
Schedule Production
60% 50% 40% 30% Item/ Location
20% 10%
Total Item All Locations
0% DFU Item Family
4
Months Before Sales
3
2
1
0
Sale
Benefits of Measuring Forecast Performance • Provides a benchmark for continuous improvement of the forecasting process. • Allows forecasters to manage by exceptions. • Measures the accuracy of both the base statistical forecast and the adjustments. Are my adjustments working? • What gets measured gets better!
Forecasting Forecast Performance Analysis How is my accuracy tracking? Is it improving over time?
Total History
Total Forecast
Total Error
