Databricks study: governed context lifts Genie One accuracy to 93–95%
Start here
Databricks published a study measuring how much Genie One's answer accuracy improves when curated agents are paired with a governed context layer. On a fictional commercial auto insurer, where Finance and Underwriting gave contradictory but correct loss-ratio answers, tables alone scored 64% accuracy; curated agents and the governed layer each reached 83% on their own, and using both together lifted accuracy to 93–95% across two runs. In a second test, adding the governed layer to curated agents raised a fictional retailer's accuracy from 54% to 82%. Databricks says Genie One, powered by Genie Ontology (now generally available), brings this business context into the analytics experience. Databricks describes the retailer benchmark as 49 questions, a smaller check of the pattern rather than a second full study, and says the work tested business context rather than access control.
Generated from reports · updated 2 h ago
Report timeline
Follow the reports to see the event from different sides.
- Databricks BlogDatabricks Genie One Accuracy Study: Why Curated Agents Need Governed Context
Databricks tested Genie One on a fictional commercial auto insurer where Finance and Underwriting gave contradictory but correct loss-ratio answers. Tables alone scored 64% accuracy; curated agents and the governed layer each reached 83%, and combining both lifted accuracy to 93–95% across two runs. The governed layer also raised a fictional retailer's accuracy from 54% to 82% when added to curated agents.
Heat for this event
Not enough continuous observations yet to draw a trend.