Automatic sales promotion selection system and method

• US 5,774,868 A
• Filed: 12/23/1994
• Issued: 06/30/1998
• Est. Priority Date: 12/23/1994
• Status: Expired due to Term

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: International Business Machines Corporation
• Inventors: Lund, Stephen Hollis, Cragun, Brian John, Kelsey, Todd Mark
• Primary Examiner(s): Weinhardt, Robert A.
• Application Number:
  US08/363,053
• Time in Patent Office: 1,285 Days
• Field of Search: 395/201, 395/210, 395/214, 395/216
• 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
• CPC Class Codes: G06Q 30/0207 Discounts or incentives, e.g. coupons or rebates
• CPC Class Codes: G06Q 30/0255 based on user history

Abstract

An automated sales promotion selection system uses neural networks to identify promising sales promotions based on recent customer purchases. The system includes a customer information device that receives customer data relating to customer purchases of items from an inventory of items, a central processing unit having a sales promotion neural network and a storage unit containing a plurality of item identifiers comprising potential customer purchases of additional items from the inventory, wherein the sales opportunity neural network responds to customer data received from the customer information device by determining if one or more of the item identifiers in the storage unit corresponds to an item likely to be purchased by one of the customers, and an output device that receives the item identifiers of the likely purchases determined by the sales promotion neural network and produces a sales promotion relating to at least one of the item identifiers.

Claims

1. An automated sales promotion selection system comprising:

a customer information device that receives customer data relating to purchases of items by customers;

a computer system including a central processing unit and a storage unit containing a purchase advisor neural network and a plurality of item numbers that identify items available for purchase, wherein the purchase advisor neural network responds to customer data received from the customer information device and to current purchase data comprising item numbers of a current purchase of a customer by determining if one or more of the item numbers stored in the storage unit corresponds to an item likely to be purchased by the customer but not present in the current purchase data and identifies a sales promotion relating to the item;

an output device that receives the item numbers of the likely purchases determined by the purchase advisor neural network and produces the identified sales promotion; and

a customer demographics neural network that estimates buying characteristics of one or more customers most likely to be at a purchase location, and also produces item numbers comprising the estimated item purchases of the estimated customers.

2. An automated sales promotion selection system as defined in claim 1, wherein the central processing unit receives the item numbers of the estimated purchases from the customer demographics neural network, segments the item numbers into purchase classes, and provides the purchase advisor neural network with the segmented item numbers as input;

and wherein the purchase advisor neural network responds to the input by determining if one or more of the item numbers corresponds to an item likely to be purchased by one of the estimated customers.

3. An automatic in-store sales promotion selection listing system that identifies items for likely purchase by a customer during a current in-store visit, the system comprising:

a customer billing terminal that generates sales purchase data comprising a sales transaction identifier, sales date, and purchase items relating to purchases of items by the visiting customer during the current in-store visit;

a data storage unit containing a set of item numbers identifying items and a set of archival sales information data comprising sales information data relating to prior purchases of items by one or more customers during prior in-store visits;

a sales promotion neural network that assigns each of the item numbers in the storage unit a value indicating the likelihood of the item being purchased by the visiting customer during the current in-store visit;

a central processor unit that provides the sales promotion neural network with sales purchase data relating to the current in-store visit by the visiting customer and selects one or more of the item numbers as most likely to be purchased by the visiting customer;

a neural network training program that provides sales information data stored in the data storage unit to the sales promotion neural network during a training process; and

a customer sales output device that generates a listing of one or more item numbers of the items selected by the central processor unit as likely purchases to be suggested to the visiting customer.

4. An automatic in-store sales promotion selection system as defined in claim 3, wherein the central processing unit further includes a neural network training program that collects a set of sales purchase data for customers, selects a training epoch subset of the collected sales purchase data, and performs a neural network training process with the selected data, and wherein the neural network training program further repeatedly collects data, selects a training epoch subset, and performs the training process until all neural network training epoch data subsets in the collected sales purchase data have been processed.

5. An automatic in-store sales promotion selection system as defined in claim 4, wherein the neural network training program collects sales purchase data comprising a purchase event identification, purchase date, time of purchase, and item number.

6. An automatic in-store sales promotion selection system that identifies items for likely purchase by a customer during a current in-store visit, the system comprising:

a customer billing terminal that generates sales purchase data comprising a sales transaction identifier, sales date, and purchase items relating to purchases of items by the visiting customer during the current in-store visit;

a data storage unit containing a set of item numbers identifying items and a set of archival sales information data comprising sales information data relating to prior purchases of items by one or more customers during prior in-store visits;

a sales promotion neural network that assigns each of the item numbers in the storage unit a value indicating the likelihood of the item being purchased by the visiting customer during the current in-store visit;

a central processor unit that provides the sales promotion neural network with sales purchase data relating to the current in-store visit by the visiting customer and selects one or more of the item numbers as most likely to be purchased by the visiting customer;

a neural network training program that provides sales information data stored in the data storage unit to the sales promotion neural network during a training process; and

a customer sales output device that generates a listing of one or more item numbers of the items selected by the central processor unit as likely purchases to be suggested to the visiting customer;

wherein the central processing unit further includes a customer demographics neural network that estimates customers most likely to be at a purchase location and their respective estimated item purchases, and wherein the sales opportunity neural network then selects item numbers as likely purchases by estimating items most likely to be purchased by the estimated customers.

7. A method of dynamically identifying sales opportunities for purchases of items by customers from an inventory of items, the method comprising the steps of:

training a purchase advisor neural network that generates an output set of item numbers comprising sales opportunities for purchases of the items;

providing the trained purchase advisor neural network with customer data;

generating a sales opportunity output with the trained purchase advisor neural network in response to the customer data, the output including one or more item numbers that identify items in the inventory, and

selecting a set of item numbers from among the sales opportunity output generated by the purchase advisor neural network as potential purchases from the inventory of items;

wherein the step of providing customer data comprises providing the purchase advisor neural network with data that relates to a purchase of one or more items by a customer that occurred during a present visit by the customer to a purchase location; and

wherein the step of selecting item numbers of potential purchases for the customer comprises the steps of;

estimating buying characteristics of one or more customers most likely to be at a purchase location; and

estimating item numbers of items most likely to be purchased by the estimated customers.

8. A method of dynamically identifying sales opportunities for purchases of items by customers from an inventory of items, the method comprising the steps of:

training a purchase advisor neural network that generates an output set of item numbers comprising sales opportunities for purchases of the items;

providing the trained purchase advisor neural network with customer data;

generating a sales opportunity output with the trained purchase advisor neural network in response to the customer data, the output including one or more item numbers that identify items in the inventory; and

selecting a set of item numbers from among the sales opportunity output generated by the purchase advisor neural network as potential purchases from the inventory of items;

wherein the step of training the purchase advisor neural network comprises the steps of;

collecting a set of sales purchase data for a plurality of customers;

selecting a training epoch subset of the collected sales purchase data;

performing a neural network training process with the selected data in which network coefficients are modified;

repeating the steps of selecting and performing until all neural network training epoch data subsets in the collected sales purchase data have been processed; and

wherein the step of providing customer data comprises the steps of;

training a demographics neural network that generates an output set of data defining predicted purchases of customers during a purchasing transaction;

providing the trained demographics neural network with prediction data comprising the current date, current time of day, and environmental information; and

generating the neural network predicted customer purchases.

9. A method of dynamically identifying inventory items having an increased likelihood of being purchased by customers during transactions, thereby comprising sales opportunities, the method comprising the steps of:

creating a purchase detail file having inventory item numbers corresponding to inventory items purchased by a customer during a current transaction time;

comparing each inventory item number in the purchase detail file against a plurality of predetermined purchase classes that define sets of inventory items that are likely to be purchased together by customers during a single transaction;

creating a class definition file having the information from the purchase detail file along with the assigned purchase class of each inventory item, for each match found in the step of comparing;

identifying inventory item numbers corresponding to inventory item numbers that are not included in the purchase detail file information for the current transaction time but that are included in one or more of the assigned purchase classes in the class definition file; and

providing an output set comprising all the inventory item numbers identified as not included in the step of identifying.

10. A method as defined in claim 9, wherein the step of creating a class definition file comprises the steps of:

creating a class definition file data structure;

determining if each purchased inventory item for a current transaction time is a member of an assigned purchase class and if the assigned purchase class has not previously been encountered in the current transaction time;

calculating an item probability corresponding to the likelihood of the inventory item being present in a purchase, defined by the frequency of being purchased at a given transaction time when at least one member of the purchase class also is purchased, and accumulating the determined probability into the class definition file data structure if the outcome of both determinations in the step of determining were affirmative;

comparing a predetermined key item threshold value for the assigned purchase class of the inventory item against the probability calculated in the step of calculating and returning to the step of determining for processing the next inventory item if the threshold value is not exceeded.

11. A method as defined in claim 10, further including the steps of:

creating a class update data structure;

creating a temporary class definition data structure for each transaction time in the purchase detail file, comprising a copy of the class definition data structure;

matching each inventory item in the transaction time to each class in the temporary class definition data structure and incrementing an accumulation count for the class with each match; and

adding the purchase transaction to the class update data structure if the accumulation count for the class exceeds a predetermined class threshold value.

12. A method as defined in claim 11, further including the steps of:

identifying each inventory item belonging to a purchase class in a transaction time of the class update data structure;

setting an item count of the identified inventory item to an initial value upon first identifying the inventory item and incrementing the item count each time the inventory item is identified in the class update structure thereafter;

identifying each inventory item belonging to a purchase class in the class definition data structure after all inventory items in the class update data structure have been identified;

calculating a new probability value corresponding to the likelihood of the inventory item being present in a purchase, defined by the frequency of being purchased at a given transaction time when at least one member of the purchase class also is purchased;

updating the new probability value into the temporary class definition data structure for the inventory item and class; and

repeating the steps of identifying, calculating, and updating for all classes in the class definition data structure.

13. A method as defined in claim 12, further including the steps of:

sorting the class update data structure according to the updated item probability values for each inventory item in a purchase class;

adding the inventory item to the class definition file data defining the purchase class if the updated item probability is greater than a predetermined class definition threshold value and if a predetermined maximum purchase class size has not been exceeded;

setting a key item flag if the updated item probability is greater than a predetermined key item threshold value;

creating a class change file data structure for the purchase class and placing into it all changes made to the class definition file; and

repeating the steps of sorting, adding, setting, and creating for each purchase class of the class definition file.