Build a sales dashboard with AI: the checklist

For Lovable, Claude or any other AI builder. Print it, tick it.
Ahmed Essam
crmmeister.com

1. Before the first prompt

  • Export a sample, not your whole CRM. Remove names, e-mails and notes you do not need.
  • Write down the 8 to 10 KPIs you will actually look at on a Monday. Cut the rest.
  • Decide the date logic: results by close date, pipeline as of today. Write it into the prompt.
  • Define win rate in one line: won / (won + lost), or won / all. Pick one.
  • Write cleanup rules for the mess you know: duplicates, empty owners, empty values, odd stage names.
  • Calculate 3 reference numbers by hand or in a spreadsheet, so you can check the AI later.

2. Prompting

  • Put the whole spec in the first prompt: who uses it, data, cleanup, filters, KPIs, design.
  • After that, one change per prompt, and say what must stay untouched.
  • Ask the tool to check its own numbers against a separate calculation, and to tell you the values.
  • Lovable: ask for a client-only app with the data bundled if you do not want a database.
  • Claude: ask for one HTML file, no external requests, plain scripts (no modules).
  • Save every prompt you send. It is your rebuild recipe.

3. Break it before you trust it

  • Compare the headline numbers with your 3 reference numbers.
  • Combine filters: one rep + one source + last quarter. Do totals still add up?
  • Pick a filter combination with no deals. Empty state, or broken chart and "NaN"?
  • Click a chart bar: does the deal list match the bar?
  • Open it on a phone (390 px) and use it with the keyboard only.
  • Export the CSV and open it in Excel: same rows as on screen?

4. Sharing and data safety

  • Know who can open the link. Public link, login, or a file on your own server?
  • Check the tool's training setting for your plan and opt out if you need to.
  • Never paste an API key into a prompt. Use the tool's secrets screen.
  • For live CRM data use a read-only key that you can revoke.
  • Dummy data for demos. Real data only where you control the hosting.

5. Next Monday

  • Test the refresh: drop in a new export. Do the numbers change, or were they hard-coded?
  • Write down who updates it and when. A dashboard nobody refreshes is a screenshot.
  • Keep the data quality panel. Fix the flagged deals in the CRM, not in the dashboard.

6. When to stop and buy

  • More than 1 data source, or live data every hour: look at your CRM's own reports or a BI tool first.
  • Many viewers with different permissions: you need logins and row rules, not a file.
  • No one on the team can read the code: keep it small, or keep it in a tool with an editor.
From "Build a sales dashboard with AI: Lovable vs. Claude"crmmeister.com