AI assortment recommender

Problem

Assortment decisions were manual, fragmented and driven by cluster-level logic. Merchandisers balanced product performance, store specifics, trends and business constraints with no centralized, scalable decision support, producing inconsistent assortments, slow refresh cycles and heavy reliance on manual analysis.

Approach

As the team lead, I drove the design and build of an assortment recommendation engine on Databricks. It unifies sales, inventory, product attributes, store clusters and demand signals into one governed data foundation, then recommends the most relevant products per store with dynamic logic: relevance from historical performance and trends, support for new-product introduction and renewal, controlled refresh to optimize turnover, and pinning to protect key products. Built for production scale, with robust pipelines, data-quality checks and monitoring.

Impact

Shifted assortment from manual definition to AI-assisted decisions, improving consistency and the balance between global coherence and local adaptation across the store network. It laid the foundation for a broader retail-AI ecosystem connecting recommendations with optimization, product strategy and end-to-end planning.

Built on Databricks.