CVS Health turns retail forecast questions into fast answers with Databricks Genie Agents

Answering why at retail scale

CVS Health's data science team forecasts demand and labor needs for its stores, and every week business partners ask why. Questions often include:

• Why were we under forecast?
• Why did sales drop?
• Was it weather, a promotion, an event or a store closure?

Each answer meant a ticket to a data scientist, who reviewed the problem store by store, tested hypotheses, wrote queries and built a report. A single request could take one to three days.

Dashboards couldn't keep up. Every user wanted a different view, and as charts piled on, the dashboards became too complex to self-serve. Data scientists ended up working like analysts, writing SQL and formatting emails instead of improving models.

Leveraging Genie Agents and Unity Catalog for insights

CVS Health built a retail-specific Genie Agent on governed data in Unity Catalog with Azure Databricks that investigates across data sources and returns a summarized answer. The team is shifting from a push model, where data scientists send reports, to a pull model, where business users ask Genie directly. Instead of 100 dashboards for 100 users, CVS Health can offer 100 perspectives from the same governed data foundation.

Accelerating ad hoc requests with Genie Agents

Turnaround for ad hoc requests has dropped significantly from the old one- to three-day window. In a demo on synthetic data, Genie completed a full why investigation in a few minutes, comparing a top store against other top performers and surfacing factors like store size and promotions.

Users can ask questions directly from a dashboard visualization or in the dedicated Genie Agent, and Genie keeps context across follow-ups. With less time spent on queries and emails, data scientists are investing more in time-series models, which has contributed to better forecast accuracy. Faster explanations have also built trust with stakeholders and partners, such as finance and merchandising.

Enhancing governance and context for Genie Agents

CVS Health started with data governance, not AI. Tables are registered in Unity Catalog, where AI-generated column descriptions are validated with domain experts to teach Genie business meaning.

The Genie Agent adds retail context on top through configuration, such as joins between tables, SQL expressions for metrics like sales per square foot, measures, filters and example SQL queries. The team has refined its retail Genie Agent for about nine months, building a library of training questions along the way. As the team puts it, Genie isn't replacing analysts. It's making high-quality analysis available to more people.