ES Loyalty Boost 2.8 release notes
These release notes cover features introduced or enhanced since ES Loyalty Boost 2.7.1.
What's in this release
- Gen AI Product Affinity Scores (enhancement)
- Stretch Configuration Enhancement (enhancement)
Enhancements
Gen AI Product Affinity Scores
Previously, product affinity prompts were triggered directly in Snowflake using OpenAI. This enhancement replaces that approach with a trigger-based method integrated with the GenAI module during offer pool uploads.
The GenAI module is now invoked in the following scenarios:
- When an offer pool is uploaded for the first time and there is no data for
GLOBAL + PRODUCT_SELECTORSin the campaign processor - When there is a difference between the old and new
PRODUCT_SELECTORSlist
Affinity score types:
| Type | Description |
|---|---|
| Member attribute product affinity | Predicts propensity to buy a product based on member attributes such as gender, average household income, province or state of residence, RFM index, and days since last purchase |
| Past purchase product affinity | Predicts propensity to buy products related to past purchases, based on a GenAI score for products that are similar or part of the same purchasing pattern — similar to "Products related to this item" on Amazon |
This feature enables comprehensive evaluation of buying patterns to provide more relevant rewards, increasing member engagement and purchase activity.
See also:
Stretch Configuration Enhancement
Previously, Loyalty Boost used a single percentage to calculate the ideal stretch value for all members across all slot types — based solely on average past spend. This uniform approach produced less effective stretches, particularly for product categories where members have natural purchase limits, such as fuel.
This enhancement calculates stretch on a per-member basis by incorporating both the average spend and the standard deviation of purchases within each product selector. If a member has fewer than a defined minimum number of transactions in a product selector, their stretch is derived from the population rather than their own history.
How the enhanced logic works:
| Member spending pattern | Stretch behavior |
|---|---|
| Spends a very similar amount each transaction (low standard deviation) | No stretch applied — consistent spend suggests the member is already at their maximum |
| Has spent a high amount at some point but usually spends low | Higher stretch applied — capturing the potential for that larger spend to recur |
| Has fewer transactions than the minimum threshold | Stretch calculated from population history to smooth the baseline |
This approach better fits stretch offers to the individual member, resulting in more effective offers and improved outcomes.
See also: