Autopilot Docs
AI features

The AI deal scanner

Photograph a paper write-up sheet — Autopilot reads the handwriting and fills the deal form.

Last updated 20 June 2026

If your salespeople still write deals on paper before entering them, the AI deal scanner cuts the data-entry step. Take a photo, AI reads the sheet, the form fills itself.

How to use it

  1. Click Add deal to open the deal form.
  2. Click Scan write-up at the top.
  3. Take a photo (phone) or upload an image (desktop).
  4. Wait ~5 seconds while AI reads it.
  5. Review what got picked up. Fix anything wrong.
  6. Hit Save.

What it reads well

  • Customer details — name, phone, address. Even messy cursive.
  • Vehicle details — make, model, year, colour. Variants too if you've written them clearly.
  • Sale price, trade-in allowance, deposit — anything dollar-formatted.
  • Dates — DD/MM/YYYY, "May 5", "5/5/26", all fine.
  • Tick boxes — manager signed / customer signed checkboxes.

What it struggles with

  • Heavily abbreviated handwriting — "T7P L AWD WHT 2024" doesn't always resolve. Spell things out where you can.
  • Carbon-copy paper that's faded — try a brighter shot or use the original sheet.
  • Multiple deals on one sheet — process them one at a time. The scanner assumes one deal per image.
  • Custom fields without clear labels — if your form has a "Paint protection sold" field, write the label clearly on the paper sheet.

What gets matched

The scanner sees your dealership's deal form schema — which fields exist, what types they are, what options dropdown fields accept. So it only fills fields you actually have on your form.

If you customise your form (add fields, rename fields), the scanner picks up the changes automatically — no retraining required.

Customer-facing vs internal mode

The form has two modes:

  • Internal (default) — every field, including cost and margin.
  • Customer-facing — financial fields hidden, used when the customer is sitting next to you.

The scanner respects whichever mode you're in. In customer-facing mode, it won't try to read or fill cost/margin fields.

Privacy

Images are sent to Google's Gemini API for OCR. They're processed in real time and not used to train models (we use the no-retention API tier). The image isn't stored after the response comes back.

If you'd rather not use AI scanning at all, just don't click the button — the form works fine for manual entry. There's no setting to disable the button entirely (yet); tell us if it's a blocker.

What happens if it gets something wrong

You're always reviewing the result before saving. The scanner pre-fills; you confirm. So a wrong reading is just a typo correction, not a save-and-discover-later problem.

If the scanner consistently mis-reads a specific field type, send us the image and the actual value — we feed those back into prompt tuning. Most "doesn't work" reports turn out to be a fix on our end.