Unifying HR data for enhanced analytics
Mars, the family-owned company behind brands like Snickers, M&M’s, Ben’s Original, VCA and Banfield, has grown through acquisitions including Kellanova, maker of Pringles and Pop-Tarts. Its people analytics team had brought data from multiple HR sources into Azure Databricks and built reusable datasets, but the data wasn’t unified within a common data model. Dashboards were siloed by domain, and business translators relied on data analysts who could spend days answering a single question.
Mars wanted an associate 360: one governed view of every associate from hire to retire that business users could explore in natural language.
Enhancing workforce intelligence with Genie Agents and Unity Catalog
Mars transitioned its people data to Unity Catalog for governance, lineage and security. Working with Tiger Analytics, the team built a common data model across recruiting, performance, compensation, core HR and learning. It included semantic views and business context so that Genie would understand Mars terminology.
Accelerating data access with Genie Agents
Business translators on the people analytics team now ask Genie the questions they used to send to data analysts and get results fast, without an analyst spending days on the request.
With the common data model spanning every HR domain, Genie Agents help users:
Ask cross-domain questions that connect recruiting, learning and performance data, beyond what siloed dashboards could answer.
Analyze results at different levels of the org hierarchy.
Ask follow-up questions in the same conversation, with Genie keeping the context.
In a demo using synthetic data, Genie interpreted a vague prompt such as “Tell me about attrition” and returned multi-year trends. It then answered follow-ups on which countries were above average and how many leavers were high performers, asking a clarifying question along the way.
Securing HR data with Unity Catalog
Genie Agents are tailored to personas such as recruiters and learning managers. Unity Catalog ensures each user sees only the data they’re allowed to access, supporting natural-language HR analytics without compromising governance.
Improving Genie Agents’ accuracy and performance
The team tested Genie against the data model while building it. When early results fell short, they added a link table that provides a time-phased view of each associate, down to the day. Genie checks the link table first, which improved accuracy and performance and reduced hallucinations.
Next, Mars plans to improve visualizations and fine-tune persona-based experiences for HR and finance business partners. It also plans to explore predictive and prescriptive insights in partnership with legal.
FAQ: Mars on Databricks Genie
Mars built Genie Agents on a common data model that unifies associate data from hire to retire. Business translators ask questions in natural language and get results fast instead of waiting days for an analyst.