Authors:
James Hawk, Patricia Hogan
Abstract:
TGCS has already provided self-service technologies that allow for product identification of non-barcode items, such as produce. This has inherent inaccuracies, so the shopper is presented with a set of items that closely match the item being purchased. For example, a shopper may be buying a Gala apple, but the produce recognition solution cannot distinguish Gala from Honeycrisp, Rome, etc. Also, there are a set of organic options that further introduce uncertainty.
For any shopper, the POS system will not know how truthful that shopper is. An unscrupulous shopper would choose the cheapest item from the suggested list.
This disclosure is intended to develop an 'honesty' profile to quantify the likelihood of that shopper selects the correct item.
Background:
The problem being addressed by this idea is figuring out how likely a shopper is to deliberately identify their produce item as a cheaper item, so that the retailer knows they should offer this shopper extra assistance to identify the correct item. This helps reduce some of the shrink that occurs in retail environments by mis-identified non-barcode items. Produce recognition is imperfect and providing a mechanism for improving the security of the POS transaction would increase the value add of the produce recognition feature.
Description:
Using loyalty data, or in the future facial recognition, the shopper can be identified -- or can be assigned as non-Personal Identification Information (PII) identifier (ID number). This ID can be used to persist data for this shopper. When the produce recognition is used, the set of choices will be seeded with an obviously incorrect, but cheaper choice. In the example above, if a Gala apple (reddish in color) is being purchased, Golden Delicious could be included in the possible matches. The system would then keep a historical data record of the obviously incorrect choices to create an 'honesty' score. Based on the honesty score of the shopper at the self-checkout, the store's attendant could provide special 'assistance'. A key dependency using prior art is being able to assign an identifier or ID number to a shopper. Presumably, this needs to be free of all PII, so a loyalty number or a simple hash of the loyalty number could suffice (call this our SecID). The SecID would be used as the index into an 'honesty' database. This 'honesty' database would also include data elements to accumulate 'correct' and 'incorrect' selections as well as an 'honesty' score.
Part 1
As part of the produce recognition system, each item has a visually different and cheaper item associated. For the cheapest items, this is not relevant, and this method does not apply. While in use, the different/cheaper item is presented, and the shoppers 'honesty' score is updated based on their selection of the obviously wrong item. If the wrong item is selected, the 'incorrect' data element would be incremented. For the correct item, the 'correct' data element would be incremented. The 'honesty' score is recalculated based on the ratio of 'correct' divided by total (incorrect + correct).
Part 2
Once a minimum data set is achieved, e.g. (incorrect + correct). > 20, the 'honesty' score is available for use. The actual minimum would be set based on the retailer's policies and operational needs. Also, a threshold for the 'honesty' score would be pre-set again with the involvement with the retailer's policies. If the shopper's 'honesty' score falls below the threshold, then a 'low honesty' attribute would be used in part 3.
Part 3
When the produce recognition is used, if there is sufficient history within the 'honesty' database, the Shopper Assistant (SA) in the store would be alerted if that shopper had a low honesty score. Depending on the retailer's preference, the SA could either observe or provide more "personalized assistance."
Claims:
The main claim of this idea is having the produce recognition system deliberately put a cheaper item but incorrect prediction in the suggestions to see if that will trigger the shopper to be dishonest.
TGCS Reference 2373
