Answer every team's data requests from a Slack channel
Eight Sleep lets anyone tag Devin in its data Slack channel, and it pulls numbers, builds data models and dashboards, and checks suspicious figures for Product, Ops, Finance, Growth and R&D.
What they did
The team gave Devin access to the dbt repository, Snowflake, Looker and Amplitude, then added curated notes for specific tasks; setup took a couple of hours. Staff post requests in Slack and Devin answers, or ships new dbt models, semantic models and dashboards. When a new daily sales dashboard showed unusually high revenue, Devin traced the Looker query back to Snowflake and found an email campaign had simply overperformed.
What happened
The company says it now ships about three times as many data features and investigations per week, and the queue of ad hoc data pull requests is near zero.
Keep in mind
The story was published on Devin's maker's blog and gives no accuracy rate or baseline for the threefold figure. Access is limited by permissions, and some parts, such as charting through a new connector, were still experiments.
Try it with your agent
You have read access to <your data warehouse> and <your BI tool>. When someone asks a data question in <your Slack channel>, answer with the number, the query you ran and the table it came from. If a figure looks unusual compared with the previous <number> weeks, trace it to the source tables and explain whether it is a real change or a data problem. Never write to production tables.
Replace the <angle brackets> with your own details.