Uber's finance team asks its data questions in Slack via the AI agent Finch
By Uber (Eddie Huang, Taya Yusuf, Tim Ross)Built and shared by Uber (Eddie Huang, Taya Yusuf, Tim Ross), not by me. Summarised from the public sources below. Check them before you rely on a number.
- Who
- Uber (Eddie Huang, Taya Yusuf, Tim Ross)
- Category
- Data & analytics
- Source
- Web
- Dates
- Published 17 Jul 2025 · Found 17 Oct 2026
Uber's FinTech team built Finch, an AI agent in Slack that lets financial analysts and accountants ask for financial data in plain language. Finch writes and runs the SQL query itself and sends the result back, with a link to Google Sheets.
What they built
A multi-agent system built on LangGraph, in which a supervisor agent routes questions to specialised sub-agents such as a SQL Writer Agent. Finch maps internal finance terms (for example 'US&C' and 'GBs') to tables and values through an OpenSearch index of metadata and natural-language aliases, and works against curated single-table data marts. Access is limited by role-based permissions, and the models are reached through Uber's internal Generative AI Gateway.
The result
Uber does not give hard numbers. Analysts no longer have to write complex SQL or wait for data requests, and each sub-agent is tested continuously against 'golden' queries.
Sources
Novo Nordisk analyst built a text-to-SQL agent that frees up half a full-time role
By Novo Nordisk (Rare Disease Insights & Analytics)
A senior analyst on Novo Nordisk's Rare Disease Insights & Analytics team built an AI assistant himself that answers business questions in plain language directly on healthcare claims and prescription data.
OpenAI's internal data agent takes employees from question to insight in minutes
By OpenAI (Bonnie Xu, Aravind Suresh, Emma Tang)
OpenAI built an internal data agent that teams in engineering, data science, go-to-market, finance and research use to answer data questions. The agent finds the right tables itself, runs SQL and writes notebooks and reports, so an analysis takes minutes instead of days.
Vercel removed 80% of its data agent's tools and made it 3.5x faster
By Vercel
Vercel had spent months building an internal text-to-SQL agent, d0, with specialised tools and heavy prompt engineering. When the team replaced most of the tools with a sandbox where the agent can run bash commands over the semantic-layer files, the agent became both more accurate and faster.
Want this in
your business?
Tell me which case caught your eye. I will tell you straight whether it fits, and what it would take.
Start a conversation →