Authors:
Patricia Hogan
Abstract:
This Disclosure proposes using machine learning to decide what items to move from unconsolidated item security to consolidated item security on eBOSS.
Background:
eBOSS is an Enterprise server that manages self-checkout lanes across multiple stores. Item security is the expected weight, height, length, width of an item. When an item is scanned at self-checkout, there are sensors in the bagging area that sense if the item placed in the bagging area matches the expected weight and/or dimensions of the item scanned. If the actual measurements do not match the expected measurements for the item, the system can send an alert that it thinks the wrong item was placed in the bagging area.
Unconsolidated item security is where each store of all the managed stores, has its own separate item security record for an item. If there are 2,300 stores, there are 2,300 security records for each item. If a retailer has 300,000 items, there will be 2,300 stores X 300,000 items item security records in the eBOSS database. This is very wasteful on space and is inefficient because the item security table is too large.
Consolidated item security is when we have one item security record per item, and we say this item security record is common for all 2,300 stores. In our implementation, if a security record is common across all stores, we use the -1 (minus one) store. This way there is only one security record for a can of Supplier A's mushroom soup. The expected measurements for Supplier A's mushroom soup are common across all stores. This is much more efficient.
Consolidated item security can be setup in the eBOSS by the retailer manually entering item code ranges for common items that should have the same measurements in every store. This could be based on Manufacturer family codes.
Most retailers start out with unconsolidated item security, because it takes effort to identify item code ranges for items that will have the same weight in every store. Retailers feel like they don't have time to identify and key in these item ranges.
Our idea is to use machine learning to identify items that have the same measurements in every store and convert the unconsolidated item security data for those items to consolidated item security data. This way the system gains the efficiency of consolidated item security, but the retailer does not have to type in item code ranges.
Description:
The method is to analyze item security records with the same item code but with different store numbers to
- see if the item measurements are the same for that item code in each store
- create a consolidated item security record for that item code, by moving the common measurements for that item to the "-1" store and deleting the item security records for that item code for any store numbers that are not "-1".
Machine learning algorithms that use clustering and partitioning techniques could be used to identify an item that has the same weight, height, length, depth in each store.
This method be applied several different ways. One embodiment would be to run the algorithm in the background all the time. Another embodiment would be to run the algorithm as part of online history analysis. Another embodiment would be to run the algorithm as part of "offline" history analysis. Yet another embodiment could be to export the item security data from eBOSS to another machine, and the analysis could be run on that machine, and the results imported to the original eBOSS machine.
History analysis is the existing eBOSS process that looks at the item security records to see if the item measurements have changed, and we need to adjust the expected item measurements. Think shrinkflation, where manufacturers are keeping the price of an item the same but shrinking the weight/volume of the item. The self-checkout machine needs to realize the expected weight for the item is getting smaller over time, and to use the new smaller weight.
TGCS Reference 3138
