Claude for Dealers / Learn / Build a Dashboard with Claude

Build Guide · Reporting You Own

Build a Dealership Dashboard with Claude from a CRM Export

Every GM has a report they have wanted for years and never gotten from a vendor. If the data exports to a spreadsheet, you can have it this afternoon.

By Gray Scott, co-founder of Claude for Dealers · Updated September 2026 · 9 min read

The six steps

Step 1: Export

Pull one to three months of records. For a lead dashboard you want: lead date, source, vehicle of interest, salesperson or BDC agent, appointment set date, appointment shown flag, sold flag, sold date. For a service dashboard: RO open date, advisor, labor hours, labor sales, parts sales, declined lines, pay type.

Step 2: Scrub

Delete the columns for customer name, phone, email, and address before the file goes anywhere. A chart of sold rate by source does not need them. If you need to follow a record through, keep the CRM record ID instead. This one habit removes most of your data risk. More in AI Guardrails for Dealerships.

Step 3: Write the five questions

Dashboards fail when they show everything. Write down the five questions you would ask in a Monday manager meeting. For example:

• Which lead sources turn into sold units, not just leads?
• Where do we lose people: set, shown, or sold?
• Who is above and below the store average on each step?
• How does this month compare to last month?
• How fast do we respond, by source and by hour of day?

Step 4: Build

For a file under a few thousand rows, upload it to Claude chat and ask for an artifact. For bigger files or something you will keep improving, use Claude Code in a project folder. The prompt is the same.

You are building a dashboard for a car dealership general manager.

DATA: The attached file is a CRM lead export. Customer PII has been removed.
First, list the columns you found and what you think each one means.
Tell me your plan and any assumptions BEFORE you write code.

IT MUST ANSWER:
1. Sold rate by lead source (sold units / leads), sorted high to low
2. The funnel: leads, appointments set, shown, sold, with rates
3. Each salesperson vs the store average on set, shown, sold
4. This month vs last month on all of the above
5. Median response time by source

RULES:
- One self-contained HTML file. Light background. Readable on a phone.
- Show the row count and date range at the top so I can verify.
- Any source with under 20 leads: show it, mark it "low volume."
- Rates to one decimal. No 3D charts, no pie charts.
- Do not drop rows silently. Tell me how many rows you excluded and why.

Step 5: Verify

A dashboard that looks right and is wrong is worse than no dashboard. Before anyone else sees it:

1. Row count. The number at the top must match the export.
2. One total. Sold units for the month must match the CRM's own report.
3. One person. Pick a salesperson and count their sold deals by hand.
4. Definitions. Ask Claude to state, in plain English, how it calculated each rate. Make sure "appointment shown" means what you mean.
5. Exclusions. Read the list of rows it dropped. Duplicates and test leads are fine. Real leads with a blank source are not.

Step 6: Refresh

Save the prompt and the five questions in a Project. Next month, drop in the new export and say "rebuild with this file." If you built it with Claude Code, ask it to make the dashboard read any file with the same columns, and the refresh becomes a file swap.

Three dashboards worth building first

When to stop and buy a tool instead

If the dashboard needs live data every hour, more than a handful of users with different permissions, or data from five systems joined together, you are describing a BI product and someone to run it. Build the simple version first anyway. It will tell you exactly what to ask a vendor for. The full framework is in Build vs Buy.

Common questions