A metrics dashboard with no analyst

The problem. You need a real view of your metrics, but you don’t have an analyst, and setting up a full BI tool for a five-person company is more infrastructure than the problem deserves.

The play. Export the raw data from wherever it lives — your database, your billing tool, an analytics platform — and have Claude turn it into a clean, readable view on a schedule you actually check.

Setup (10 min). Get your real data into a file Claude can read: a CSV export, a spreadsheet, or a connector to the actual tool if one exists. Write or update METRICS_DEFINITIONS.md with your exact definitions first — this is where most dashboards quietly go wrong.

The prompt:

Here's our raw data export: [paste or point at the file]. Here's
METRICS_DEFINITIONS.md for exactly what each term means at our company.

Build a one-screen summary:
1. The 4-6 numbers that actually matter, using our exact definitions
2. Trend versus last period for each — up, down, flat
3. Anything that moved more than expected, flagged with a plain-language
   guess at why (labeled as a guess, not a fact, unless you're sure)

If any number in the raw data looks inconsistent or missing, say so rather
than silently working around it.

What you get. A real, readable metrics view built from your actual data and your actual definitions — not a generic dashboard template that doesn’t know what “active user” means at your specific company.

What to watch for.

• This is only as good as METRICS_DEFINITIONS.md. Vague definitions produce a dashboard that looks precise but means something slightly different every time you check it.
• Don’t let a “guess at why” get treated as a confirmed explanation once it’s on the page — check anything surprising before you act on it.

Make it repeatable. Re-run it on the same cadence you actually check metrics — weekly or monthly, whichever matches your real habit — using a fresh export each time. Feed the same output into a one-screen board update whenever you need one.