The business situation
Product and operations teams needed faster access to data, but routine questions still depended on analysts to translate business language into SQL and visual summaries.
Back to professional experience
Natural-language analytics model for business users
PRISM, formerly OYO
As Product Manager III for the OYO US consumer platform, call centre, and G6 platform, built a two-agent natural-language-to-SQL workflow that generates SQL, executes queries, and visualizes outputs for business teams.
- Role
- Product Manager III
- When
- November 2025 - Present
- Product area
- AI PM, Data Products, SQL
The business situation.
Make operational and product data easier for business teams to query without waiting on analyst translation.
Evaluation result
85%
Accuracy against gold-standard natural-language business queries
Product decision
2-agent
Separated SQL generation from query execution and visualization
Primary users
Business teams
Product and operations teams exploring data through natural language
Accuracy was evaluated against known-good answers for a gold-standard set of business queries. Query examples, benchmark size, adoption data, and internal dashboards are confidential.
The decision I owned
I separated SQL generation from execution and visualization, so the product could measure generated-query quality and give users a clearer path to inspect the answer.
How I made the workflow trustworthy
Discovery centered on repeated business queries, common SQL patterns, data-table ambiguity, and the moments where teams waited for analyst support.
The delivery tradeoff
The product bet was to make analytics conversational while still measurable: evaluate generated SQL against known-good answers, separate generation from execution, and make output quality visible.
What shipped
The workflow used a two-agent architecture to generate SQL, execute queries, and return visual outputs, with quality measured against known-good answers. In parallel, I automated PM workflows for research, marketing benchmarks, daily Teams reporting, and PRD creation using Cowork.
What changed
The natural-language-to-SQL workflow reached 85% accuracy against gold-standard business queries and gave business users a route to explore product and operational data with less analyst dependency.
What I would carry forward
For AI products, a useful answer is not enough. The evaluation loop, known-good comparisons, and a legible path to inspect output are part of the product itself.