Challenge
The retailer’s existing product targeting model could not keep pace with weekly promotional activity. It struggled to account for customer behaviour, campaign context, and practical constraints such as budget, allergens, and store availability.
That left marketers with low-precision targeting, slower campaign setup, and limited visibility into whether offers were creating real sales uplift.
Approach
Unlocq.ai built a recommendation engine with two models working together:
- Collaborative filtering to generate customer-product affinity scores from 52 weeks of transaction history
- A contextual LightGBM model to factor in offer, product, customer, and seasonal signals
We paired those models with BigQuery, Google Cloud Composer, and Vertex AI pipelines so scoring and allocation could run as a repeatable weekly process instead of a manual exercise.
The implementation also carried the operational constraints the client needed in production, including budget caps, allergen rules, and store-level availability.
Results
- 2x higher redemption rate in test campaigns
- $700K in weekly incremental sales from boost offers
- 95% reduction in manual allocation work
- 3x faster campaign turnaround