An AI data agent that cut analysts’ time on ad hoc requests by 80%
A US insurance company’s data team spent much of its week answering one-off data requests. BetterBrain built an agent that writes the SQL and Python, checks its own work, and asks when unsure. Analyst time on these requests fell 80%.
Less analyst time on ad hoc data requests
Faster data pulls for non-technical users
BetterBrain brings together expertise from top institutions and innovative companies, combining deep academic AI knowledge with practical industry experience.
About the client
The client is a US enterprise insurance company with a data team that answers requests from across the business.
The challenge
Much of the data team’s time went to ad hoc requests. A manager needs one number, a marketer needs one list, and many of the questions had been answered before.
People outside the team could not pull data themselves, so they waited in the queue, and analysts spent their days on repeat work instead of analysis.
The solution
BetterBrain built an AI agent that takes a request in plain English and works through it the way an analyst would.
- It plans the steps, then writes and runs the SQL and Python that answer the question.
- It reflects on the result and checks its own work before showing it.
- When a request is unclear, it asks the user a question instead of guessing.
- It learns from previous queries, finding similar past requests with keyword search (BM25) and vector embeddings.
A person stays in the loop on every answer, and the interface shows what the agent did so users can trust the result. Code generation and reasoning run on OpenAI’s o1-mini.
Impact
80% less analyst time on ad hoc requests
Analysts spend 80% less time on one-off data requests, and more of it on analysis.
75% faster for everyone else
Non-technical users pull data 75% faster, without waiting in the data team’s queue.
