Sales history decomposition

• US 7,848,946 B2
• Filed: 01/12/2005
• Issued: 12/07/2010
• Est. Priority Date: 01/12/2004
• Status: Active Grant

• Current Assignee: Blue Yonder Group, Inc. (Panasonic Holdings Corporation)
• Original Assignee: JDA Software Group Incorporated (Panasonic Holdings Corporation)
• Inventors: Sabnani, Vikas, Acharya, Suresh
• Primary Examiner(s): Boswell; Beth V
• Assistant Examiner(s): SANTIAGO, LUIS F
• Application Number:
  US11/033,487
• Publication Number:
  US 20050256759A1
• Time in Patent Office: 2,155 Days
• Field of Search: 705/7, 705/8, 705/36.R, 705/10
• US Class Current: 705/7.31
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities

Abstract

A statistical system and method for filters sales history to yield two demand series—one that is time insensitive (i.e. function of price and promotional activities only) and another that is price & promotional activities insensitive (i.e. function of temporal factors only). Once this task is completed, then specialized econometric techniques can be used to model the former and specialized time series techniques can be used to model the latter. In one embodiment, the present invention provides method and related system comprising the iterative steps of:

• • mapping definitions of sales data;
  • time slotting of the sales data;
  • aggregating the sales data;
  • initial estimating of a trend in the sales data;
  • computing de-trended sales history using the trend estimate;
  • regressing to remove price, seasonal, and marketing instruments effects;
  • removing the estimated seasonal, price and marketing instrument effects from the sales data;
  • re-estimating trend effects from the sales data having the removed estimated seasonal, price and marketing instrument effects; and
  • computing a de-trended and de-seasonalized sales data.

Claims

1. A computer implemented sales history decomposition method comprising:

mapping sales data related to demand forecast units to a plurality of demand forecast unit groups for one or more product categories, wherein each demand forecast unit is a unique product identifier code, wherein the sales data relates to actual sales of products, wherein each demand forecast unit is assigned to a demand forecast unit group included in the plurality of demand forecast unit groups wherein demand forecast unit groups correspond to different categories of demand forecast units;

after said mapping, time slotting of the sales data;

after said time slotting, aggregating the sales data contained in the plurality of demand forecast unit groups;

after said aggregating, initial estimating of a trend in the aggregated sales data, wherein said initial estimating includes computing a moving average;

computing using a computer processor a de-trended sales history using the trend estimate;

after said computing, using a computer processor regressing the de-trended sales history as a function of price, seasonal, and marketing instrument effects;

after said regressing, removing estimated price, seasonal, and marketing instrument effects from the sales data using the results of regressing the de-trended sales history;

after said removing, re-estimating trend effects from the sales data from which the estimated price, seasonal, and marketing instrument effects have been removed;

computing de-trended and de-seasonalized sales data using the re-estimated trend effects and the sales data from which the estimated price, seasonal, and marketing instrument effects have been removed; and

outputting the computed de-trended and de-seasonalized sales data;

the method further comprising repeating the steps of;

computing using a computer processor de-trended sales history using the trend estimate;

regressing the de-trended sales history as a function of price, seasonal, and marketing instrument effects;

removing estimated price, seasonal, and marketing instrument effects from the sales data for the particular product using the results of regressing the de-trended sales history; and

re-estimating trend effects from the sales data for the particular product from which the estimated price, seasonal, and marketing instrument effects have been removed, wherein aggregating the sales data includes aggregating price ratios, wherein the denominator of the price ratios is an exponentially smoothed average of historical observed prices, and wherein time slotting of the sales data includes computing a reference price.

2. The computer implemented sales history decomposition method of claim 1, further comprising computing a mean revision in trend.

3. The computer implemented sales history decomposition method of claim 1 further comprising:

selecting a subset of demand forecast unit groups for analysis of trend and seasonality.

4. The computer implemented sales history decomposition method of claim 3, wherein the selected subset of demand forecast unit groups does not include demand forecast units associated with an incomplete sales history.

5. The computer implemented sales history decomposition method of claim 3, wherein the selected subset of demand forecast unit groups consists of functionally homogenous products.

6. The computer implemented sales history decomposition method of claim 3, wherein the selected subset of demand forecast unit groups consists of homogenous stores.

7. The computer implemented sales history decomposition method of claim 3, wherein the selected subset of demand forecast unit groups consists of homogenous trend and seasonalities and price sensitivities.

8. The computer implemented sales history decomposition method of claim 1, wherein the demand forecast units assigned to demand forecast unit groups relate to functionally homogenous products.

9. The computer implemented sales history decomposition method of claim 1, wherein the demand forecast units assigned to demand forecast unit groups relate to homogenous stores.

10. The computer implemented sales history decomposition method of claim 1, wherein the demand forecast units assigned to demand forecast unit groups are homogenous in trend/seasonalities and price sensitivities.

11. The computer implemented sales history decomposition method of claim 1, wherein the initial estimating of a trend includes computing an n week moving average.

12. The computer implemented sales history decomposition method of claim 11, wherein n is equal to a number of time periods in a season cycle.

13. A computer system, comprising:

a data repository formed in memory, the data repository containing sales data in the form of demand forecast units;

an application repository formed in memory containing at least a first module operable by execution on a processor to;

map said sales data to demand forecast unit groups, wherein the sales data identifies products by demand forecast unit, and wherein the demand forecast unit groups correspond to different categories of demand forecast units;

time slot the sales data;

aggregate the sales data contained in the demand forecast unit groups;

estimate a trend in the aggregated sales data;

first compute a de-trended sales history using the trend estimate;

first regress the de-trended sales history as a function of price, seasonal, and marketing instrument effects;

first remove estimated price, seasonal, and marketing instrument effects from the sales data, using the results of the regressing the de-trended sales history;

first re-estimate a trend in the sales data from which the price, seasonal, and marketing instrument effects have been removed;

second compute a de-trended sales history using the trend estimate;

second regress the de-trended sales history as a function of price, seasonal, and marking instrument effects;

second remove estimated price, seasonal, and marketing instrument effects from the sales data, using the results of the regressing the de-trended sales history;

second re-estimate a trend in the sales data from which the price, seasonal, and marketing instrument effects have been removed;

compute de-trended and de-seasonalized sales data using the second re-estimated trend and the sales data from which the estimated price, seasonal, and marketing instrument effects have been removed; and

an output device, operable to output the computed de-trended and de-seasonalized sales data in a human perceptible format;

wherein aggregating the sales data includes aggregating price ratios, wherein the denominator of the price ratios is an exponentially smoothed average of historical observed prices, and wherein time slotting of the sales data includes computing reference price.