Promotional Recommendations

Improved retail promotional targeting using multi-task recommendation models.

Key Takeaways:

Promotional content lacked personalization

Inconsistent metadata and fragmented content limited the effectiveness of promotional recommendations.

We upgraded recommendation infrastructure

A new model was deployed to personalize promotions across channels and objectives.

Multi-task recommendation modeling

A Two-Tower TensorFlow model was implemented to jointly optimize click-through and conversion signals using both long-term and short-term user behavior.

Significant gains in recommendation accuracy

Improved ranking quality translated into more effective promotional targeting.