AI data analysis: from a CSV export to a leadership dashboard
By Steven Van ·
Attach a CSV to a Bearly chat, get a checked answer from Python, then turn it into a shareable one-page dashboard.
Bearly can now take a CSV or spreadsheet attached to a chat, run Python on it to answer a business question, and then turn the result into a one-page dashboard. The workflow is built on Code Interpreter, which the Bearly docs describe in Analyze data with Code Interpreter.
The worked example
Bearly's test file was a year of orders from Quillmoor Cycles, a made-up bike company: 2,273 rows with date, region, sales channel, product category, units and revenue. Asked which region grew fastest from the first quarter to the fourth, and why, Opus 5.5 ran Python on the file twice and returned a table of quarterly revenue by region.
- Asia Pacific grew fastest, from $30.5K in Q1 to $58.7K in Q4, up 92%.
- North America and Europe were roughly flat, and Latin America fell 29%.
- The Asia Pacific gain came from more orders (84 a quarter to 147) rather than bigger ones (average order up about 10%). Bikes added $15.7K and components $10.0K. The online channel added $17.5K of the $28.2K gain.
Bearly says it checked each figure against the file with its own script and they match. The reply also raised caveats unprompted: Asia Pacific is still under a fifth of Q4 revenue, bike orders are large enough to make the category lumpy, and the export can't show underlying causes such as marketing or pricing.
From answer to dashboard
A follow-up asked for a leadership-meeting dashboard as a Page. The Page opens with a summary, then shows a tile per region (full-year revenue, share of total, Q1-to-Q4 change), a line chart with Asia Pacific in orange, and the Asia Pacific categories as bars. It appears beside the chat and can be downloaded as a single HTML file or shared.
The first version came out with blank tiles, chart and category panel, because a script the model wrote into the Page failed. One line asking for a fix produced a version with the numbers written into the page itself. Bearly's advice is that a Page is code the model writes, so it should be checked before it is sent anywhere.
Using Code Interpreter
- Attach files with the plus icon or by dragging them into the chat. Code Interpreter cannot see files that are not attached.
- Bearly can choose Code Interpreter automatically. To require it, type / in the composer and select Run Python under Code.
- A Run Python step can be expanded to show its Input and Result, or Error if the run fails. Generated files can be previewed, downloaded, or found later under Browse files.
- Each run starts with a fresh workspace, so a later run can reuse an earlier output only if that file is supplied, for example by naming it. A single run returns up to five output files.
- A team administrator can disable the Interpreter, in which case Run Python does not appear in the command menu.
