Merchandising Intelligence That Merchants Trust: Albertsons' Structure-First AI for Smarter Promotions

Retail promotion decisions involve billions of dollars and millions of interdependencies — yet most AI systems produce recommendations that merchants cannot explain, validate, or trust. Albertsons built a different kind of merchandising intelligence. Using 70 billion rows of transaction data processed on the Databricks Lakehouse, they modeled item and promotion relationships as reusable graph structures — what they call Promo Molecules — that explicitly surface demand shifts including substitution effects, halo effects, and cannibalization. These structures are not black boxes. They expose deterministic, merchant-validated signals that can be audited and acted on with confidence. Foundation models then reason probabilistically to answer natural-language what-if questions, enabling merchant-trusted promotion decisions at scale without manual overrides. In this session, Karthik Iyer explains the structure-first AI architecture behind Albertsons' merchandising intelligence platform, the business outcomes it is driving, and why explainability — not just accuracy — is the key to unlocking AI adoption in retail. This is a session for CMOs, CDOs, and retail executives who are done compromising between AI capability and business trust.