Improve Demand Forecasts by Leveraging the Retailer s

1
Improve Demand Forecasts by Leveraging the
Retailers Forecasting and Replenishment System

• James Ratliff
• Consultant
• Observed Demand
• (former Inventory Manager, Kmart)
• http//www.linkedin.com/in/jamesratliffprofile

2
Abstract

• Retailers are demanding that vendors help
  streamline inventory at both the distribution
  center and store level. Usually, the retailers
  forecasting and replenishment system generates
  orders based on the observed demand which is
  current deseasonalized demand plus trend plus
  predictable seasonal fluctuations. For several
  years now, INFOREM (Inventory Forecasting and
  Replenishment Module) has been the replenishment
  application of choice by many retailers, such as
  Walmart, Target, Kmart, Safeway, etc.
  Understanding key INFOREM forecasting concepts
  will lead to the development of more accurate
  vendor demand forecasts. During this
  presentation learn the important concepts that
  drive INFOREMs forecasting process.

3
Introduction

• Why even discuss INFOREM specifically?
• Some of the largest retailers use INFOREM to
  replenish hundreds of stores. Below is a short
  list of retailers including 2008 Sales and number
  of stores
• Wal-mart 405.607 billion 7,262 stores
• Target 64.948 billion 1,591 stores
• Safeway 44.104 billion 1,743 stores
• Macys 24.892 billion 853 stores
• Kmart 18.0 billion (Est.) 1,360 stores (Est.)
• Kohls 16.389 billion 929 stores
• Total Combined 2008 Sales 573.94 billion
• Total Combined Stores 13,738 stores
• Note Number of stores is from Store Magazine
  Top 100 Retail List 2008 published in its July
  2008 issue at http//www.stores.org/pdf/08TOP100.p
  df.

4
Agenda

• Background
• Forecasting Concepts
• Ordering Concepts
• How can the retail vendor apply these concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

5
Background

• Merchandise Planning Allocation
• Long Term Demand Supply Planning
• Sales, GM, and Inventory Turns Goal Driven
• Medium Term Demand Supply Planning
• Review Monthly (Enter New Orders Monthly)
• Production and/or Import Lead Time Driven
• Short Term Demand Supply Planning
• Review Weekly (Enter New Orders Weekly)
• Ship Lead Time Driven
• ATS commitments and STA Dates

6
Background

• Goal is to improve Forecast Accuracy
• Forecast Error (MAPE) with One-Month Lag
• Industry Range Median
• Bulk Chemicals 24 to 10 11
• Consumer Goods 40 to 14 26
• High Tech 45 to 4 28
• Source Lora Cecere, Debra Hoffman, Roddy Martin,
  and Laura Preslan, AMR Research, AMR Benchmark
  Analytix data in The Handbook for Becoming
  Demand Driven, July 2005.

7
Background

• INFOREM
• Acronym for Inventory Forecasting and
  Replenishment Module
• Created by IBM in the 1970s
• Probable offspring of IBMs 1966 IMPACT
  Inventory Program Control Techniques
• INFOREM Purchased by i2 Technology in 2000
• Update JDA Acquires i2 Technologies January,
  2010 JDA owns rights to INFOREM

8
The INFOREM system
Retailer Item Records
Build Transfer Record
Retailer Data
Outputs
Batch Processing Steps
9
Agenda

• Background
• Forecasting Concepts
• Ordering Concepts
• How can the retail vendor apply these concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

10
The INFOREM Forecast Formula

• DD X BI Sales Forecast
• DD is Deseasonalized Demand
• BI is the Base Index

11
The INFOREM Forecast Formula

• Deseasonalized Demand (DD) is INFOREMs technical
  name for an items average rate of sales per week
• The DD is dynamic in that it adjusts to meet
  retail trends experienced at the store level
• The Base Index is a number that represents the
  level of sales expected in a given week compared
  to the average
• The Base Index is not dynamic and has to be
  manually changed periodically

12
The INFOREM Forecast Formula
13
The INFOREM Forecast Formula

• DD X BI Sales Forecast
• DD is Deseasonalized Demand
• BI is the Base Index

14
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
Step 3 VRS
Input / Output
Output
Step 4
15
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
Step 3 VRS
Input / Output
Output
Step 4
16
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
Step 3 VRS
Input / Output
Output
Step 4
17
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
Step 3 VRS
Input / Output
Output
Step 4
18
Step 3 DD Update Calculation

• Deseasonalized Demand Update (DDU) process
  performs three primary functions in creating a
  new DD
• Deseasonalizes current period sales (CPS)
• Calculates forecast error
• Calculates a new DD

19
Step 3 DD Update Calculation

• Deseasonalizes current period sales (CPS)
• DD Update process begins by deseasonalizing the
  current period sales (CPS).
• Dividing CPS by the Base Index (BI) for that week
  to establish a current DD (CDD)

20
Step 3 DD Update Calculation

• Calculates forecast error
• Apply a weighting ? (Alpha) to both the old DD
  and CDD to calculate the new DD
• ? is a product of forecast error
• DD Update uses a Tracking Signal to determine
  when ? needs to be increased or decreased

21
Step 3 DD Update Calculation

• Calculates forecast error

MSDTSt MADTSt
Tracking Signal TSt
where .1 lt TS lt .9 and denotes absolute value

• MSDTSt Mean Signed Deviation for Tracking
  Signal (mean forecast error)
• MADTSt Mean Average Deviation for Tracking
  Signal (mean absolute forecast error)

22
Step 3 DD Update Calculation

• Calculates forecast error

• MSDTSt ? (CDDt old DDt) (1 - ?)
  MSDTSt-1
• MADTSt ? CDDt old DDt (1 - ?)
  MADTSt-1
• ? Beta, smoothing constant often 0.1 or 0.2

Trigg, D.W. Monitoring a Forecasting System.
Operational Research Quarterly15, 1964, pp.
271-74.
23
Step 3 DD Update Calculation

• Calculates forecast error
• Damping factor (DAMP) is an assigned value from
  .1 to .9 that modifies the Tracking Signal (TSt)
  used in producing the smoothing rate ?t
• ?t is set to equal the TSt times DAMP for each
  update value from .1 to .9
• ?t TSt x DAMP

Trigg DW, Leach AG. 1967. Exponential smoothing
with an adaptive response rate. Operational
Research Quarterly 18 53-59.
24
Step 3 DD Update Calculation

• Calculates a New DD
• Variable Response Smoothing (VRS) or Adaptive
  Response Rate Exponential Smoothing (ADRES) or
  (ADRRES) is used to calculate the New DDt1
• New DDt 1 ?t CDDt (1 ?t) old DDt

25
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
DD Change Restrictions
Step 3 VRS
Input / Output
Output
Step 4
26
Calculating a New DD
Output
No Update
No Update
Step 2
Input
Step 1
Demand Filter Limits
No Update / Update ?
No Update / Update ?
Update
Update
Final New DD
New DD
DD Change Restrictions
Step 3 VRS
Input / Output
Output
Step 4
27
Summary of Steps to Calculating a New DD

• Step 1 Bypass codes are used to qualify the
  incoming sales data
• Step 2 Demand Filters are used to audit the
  sales data for reasonableness
• Step 3 Deseasonalized Demand Update (DDU)
  calculates a new DD
• Step 4 DD Change Restriction puts in place
  limits to regulate the up and down movement of
  the DD

28
Agenda

• Background
• Forecasting Concepts
• Ordering Concepts
• How can the retail vendor apply these concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

29
The INFOREM system
Retailer Item Records
Build Transfer Record
Retailer Data
Outputs
Moment of Truth
Batch Processing Steps
30
Order Forecasting Topics

• Order Paths
• Order Generation Rules
• Calculating Order Quantities
• Safety Stock

31
Order Paths
Group of SKUs
Independent SKU
JIT
Vendor Direct
Non-Flow
Flow
Non-Aggregate
Aggregate
Aggregate
Store Orders
DC Orders
Store Orders
DC Orders
DC Orders
32
Order Paths
Group of SKUs
Independent SKU
JIT
Vendor Direct
Non-Flow
Flow
Non-Aggregate
Aggregate
Aggregate
Store Orders
DC Orders
Store Orders
DC Orders
DC Orders
33
Order Paths
Group of SKUs
Independent SKU
JIT
Vendor Direct
Non-Flow
Flow
Non-Aggregate
Aggregate
Aggregate
Store Orders
DC Orders
Store Orders
DC Orders
DC Orders
34
3Tier Distribution System
Tier 1
Order Fulfillment Path
Tier 2
Tier 3
Order Fulfillment Path
Stores
Legend
National DC
Regional DCs Tier 2
Regional DCs Tier 3
35
Order Forecasting Topics

• Order Paths
• Order Generation Rules
• Calculating Order Quantities
• Safety Stock

36
Order Generation Rules

• INFOREM provides six order generation rules
• An order generation rule is a set of rules used
  to determine when to order and how much to order
• The rules are called Forecast Types and are
• Minimum / Maximum FT U
• Order Point / Order Up To Level FT O
• Presentation Stock / Slow Seller FT J
• Time Supply FT T
• Fixed Order Quantity FT F
• End of Season FT E

37
Order Generation Rules

• INFOREM provides six order generation rules
• An order generation rule is a set of rules used
  to determine when and how much to order
• The rules are called Forecast Types and are
• Minimum / Maximum FT U
• Order Point / Order Up To Level FT O
• Presentation Stock / Slow Seller FT J
• Time Supply FT T
• Fixed Order Quantity FT F
• End of Season FT E

38
Calculating Order Quantities

• The Forecast Type chosen will determine
• The Order Point
• When to order?
• The Order Quantity
• How much to order?

39
Order Forecasting Topics

• Order Paths
• Order Generation Rules
• Calculating Order Quantities
• Safety Stock

40
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
41
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
42
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
43
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
44
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
45
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
46
Calculating Order Quantities
Fixed Time Period Reordering
Idle State - Waiting for Demand
Loop 1
Demand occurs - Unit withdrawn from inventory
Loop 2
Has review time arrived?
Yes
No
Compute Inventory Status Avail OH OO - CMSTK
No
Is Avail lt OP?
Compute Order Point (OP)
End
Yes
Compute OUTL
Compute OQ
End
47
Calculating Order Quantities

• OP/OUTL Forecast Type O
• Order Point Definition
• OP FCST(RTFRT LT) SSTOCK or CSTOCK
• (which every is greater)
• When to order?
• If AVAIL lt OP
• AVAIL (Available Inventory) OH (On Hand) OO
  (On Order) - COMSTK (Committed Stock)
• COMSTK refers to store orders that are committed
  for various reasons i.e. Promotional Allocation
• RTF (Review Time Factor) - A default value of .5
  is usually used
• RT (Review Time) - Number of days between INFOREM
  reviews
• LT (Lead Time) - Number of days between order
  review and goods being available for sale ( store
  level) or distribution (DC level)
• SSTOCK (Safety Stock) - Amount of merchandise
  kept to protect against out of stock condition
• CSTOCK (Counter Stock) - The minimum amount of
  product you want to have on hand when the next
  order arrives (presentation)

48
Calculating Order Quantities

• OP/OUTL Forecast Type O
• Order Up To Level Definition
• OUTL FCST(LT OSTRAT) SSTOCK or CSTOCK
• (which ever is greater)
• How much to order?
• OQ OUTL AVAIL (rounded to pack)
• LT (Lead Time) - Number of days between order
  review and goods being available for sale ( store
  level) or distribution (DC level)
• OSTRAT (Order Strategy) - Number of days an order
  quantity should last upon receipt
• SSTOCK (Safety Stock) - Amount of merchandise
  kept to protect against out of stock condition
• CSTOCK (Counter Stock) - The minimum amount of
  product you want to have on hand when the next
  order arrives (presentation)
• OQ (Order Quantity)
• AVAIL (Available Inventory) OH (On Hand) OO
  (On Order) - COMSTK (Committed Stock)

49
Order Forecasting Topics

• Order Paths
• Order Generation Rules
• Calculating Order Quantities
• Safety Stock

50
Safety Stock

• Safety Stock is an additional layer of inventory
  added to a store order point to cover forecast
  variance to actual sales in order to prevent
  store out of stocks
• The Mean Absolute Deviation for Safety Stock
  (MADSS) is the average difference between
  forecasted demand and actual sales
• Safety Stock is controlled by forecast errors,
  service level objectives, and lead time variations

51
Agenda

• Background
• Forecasting Concepts
• Ordering Concepts
• How can the retail vendor apply these concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

52
Demand Planning Simulation
Order Paths
Group of SKUs
Independent SKU
JIT
Vendor Direct
Non-Flow
Flow
Non-Aggregate
Aggregate
Aggregate
Store Orders
DC Orders
Store Orders
DC Orders
DC Orders
53
Demand Planning Simulation
3-Tier Distribution System
Tier 1
Order Fulfillment
Tier 2
Tier 3
Order Fulfillment
Stores
Legend
National DC
Regional DCs Tier 2
Regional DCs Tier 3
54
Demand Planning Simulation
3-Tier Distribution System
Tier 1
Order Fulfillment
Tier 2
Tier 3
Order Fulfillment
Stores
Legend
National DC
Regional DCs Tier 2
Regional DCs Tier 3
55
Demand Planning Simulation
Flow Aggregate DC Orders

• How are store orders calculated?
• Store need is forecasted for each store
• All store orders are aggregated at DC level
• When product is receipted at DC, INFOREM
  calculates new OPs and OUTLs
• Product is distributed to stores based on
  reallocation process that uses the new OPs and
  OUTLs
• Reallocation process prioritizes stores

56
Demand Planning Simulation

• Spreadsheet Simulation
• Key assigned values and weekly calculated values
  in the Aux File and Policy File at the store
  level and/or DC level that some retailers provide
  through its internet portal
• - Deseasonalized Demand (DD) - Counter Stock
  (CSTOCK)
• - Base Index (BI) - Safety Stock (SSTOCK)
• - Lead Time (LT) - Order Point (OP)
• - Review Time (RT) - Order Up to Level (OUTL)
• - Order Strategy (OSTRAT) - Service Level (DSER)
  
• - Available Inventory (AVAIL) - Committed Stock
  (COMSTK)

57
Demand Planning Simulation
Using the systems forecast, replenishment, and
order settings, week one DD is calculated and a
13 week forecast for sales, inventory and orders
are calculated as shown in the spreadsheet below
and ...
58
Demand Planning Simulation
the spreadsheet chart below.
59
Demand Planning Simulation
After 13 weeks of calculating a stabilized new DD
every week, the replenishment settings begin to
destabilize due to out of stocks. Clearly, this
store did not have enough weeks of supply.
60
Demand Planning Simulation
Adding safety stock or counter stock would
probably stabilize the system. How much safety
stock or counter stock do you add to the store?
61
Agenda

• Background
• Forecasting Concepts
• Ordering Concepts
• How can the retail vendor apply these concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

62
Communicating with the Retailers Replenishment
Team
Causes of Forecast Error There are many internal
and external factors that have an impact on
forecast error.

• Bad Profile
• Improper INFOREM settings
• Lead Time Variance
• Unit Integrity

• Poor Sister Item
• Product Availability
• External Market Conditions
• Merchandise Display

63
Agenda

• Background
• INFOREM Forecasting Concepts
• INFOREM Ordering Concepts
• How can the retail vendor apply these INFOREM
  concepts?
• Demand Planning Simulation
• Working with the retailers replenishment team

64
End

• Email me at james.ratliff_at_observeddemand.com
• Or Visit my website at www.observeddemand.com
• Or call me at 312-330-2889
• Check out my profile at LinkedIn.com
• http//www.linkedin.com/in/jamesratliffprofile