Assessing damage photos with AI in 2026: how it works and what it costs

What AI sees on a damage photo

Modern AI models can read a photo the way they read text. They are trained on millions of images and therefore recognise car parts, materials and damage patterns. What an assistant extracts from a damage photo:

• The part: left door, rear bumper, wing mirror, windscreen. Even on a close-up, as long as some of the surroundings are visible.
• The type of damage: a dent without paint damage, a scratch down to the primer, a cracked bumper corner, a broken lamp.
• The extent: an estimate of the size relative to known parts, such as a door handle or a number plate.
• Whether there is more going on: does the damage touch several parts, is the dent on a crease or a seam, is something out of line.
• The quality of the photo: too dark, too much reflection, too far away. Then the assistant asks specifically for a new photo, for example from an angle in daylight.

What AI does not see

A photo shows the outside, and only the outside. Four things stay with a person:

• Hidden damage. A bumper showing a small crack can hide a bent beam or a broken sensor on the inside. You only see that on the lift.
• Repair or replace. Whether a dent can be pulled out or the panel has to come off depends on the location, the material and the paint thickness. The assistant can give a hunch, the repairer decides.
• The final price. The assistant can indicate the type of repair; the amount belongs to the quote the business makes itself, with its own hourly rate and parts prices.
• Insurance and liability. What the insurer covers and who pays is not a question for a photo assistant.

Anyone setting this up properly therefore lets the assistant give a first estimate, not a promise. The customer is told right away that it is an indication and that the business assesses the photo before an amount follows.

How it works in practice

Picture a body shop on a Saturday afternoon. A customer sends a photo of a dent in the tailgate through WhatsApp, asking whether it can be fixed. The assistant replies within seconds: it sees a dent about the size of a hand on the tailgate, without visible paint damage, and asks for a second photo from an angle plus the number plate or the car model. With that information it prepares the file: two photos, the description, the car model and a note that the dent may be repairable without respraying. The customer is told the business will call on Monday morning with a price, or can pick a moment to drop by right away.

On Monday morning the repairer does not see a loose picture in a WhatsApp conversation but a completed file. They assess the photos themselves, call the customer and make the quote. The difference is not in the judgement, which remains theirs, but in the preparation: the questions have been asked, the photos are usable and the customer already had an answer instead of two days of silence.

The same works outside the automotive trade. An installer has a photo of a leaking boiler assessed, a window company a photo of a rotten corner, a wholesaler a photo of a damaged delivery. The question is always the same: what is it, how bad is it and what is the next step.

What it costs

Photo assessment is rarely a stand-alone product. It is part of a chat or WhatsApp assistant that also answers the ordinary questions and books appointments. The price therefore follows the assistant: it depends on what it needs to know, which systems it has to use (planning, quotes, WhatsApp) and whether it also has to answer the phone. At the blue thing you get a fixed price for a first, clearly scoped version after a free introductory call. Maintenance and updates are optional afterwards from €195 per month, cancellable monthly.

The running cost of the assessment itself is small: the AI model charges a few cents per photo. At a hundred photos a week that is a few euros a month in compute. The gain is in time: no more calling customers about photos, no bad photos going back and forth twice, and a customer who already gets an answer on Saturday.

Want to know first whether it pays off for you? The Work Scan (€495 excluding VAT) calculates how much time assessing and following up on photos costs now and what an assistant takes off that. If you then have us build it, that amount is credited in full.

Ground rules for customers' photos

A damage photo belongs to the customer and often contains more than the damage: a number plate, a house number, sometimes a person. Four rules that belong in every set-up:

• Tell the customer that an assistant assesses the photo and that the business makes the final assessment. No surprises afterwards.
• Keep only what the file needs and no longer than needed. A photo that has no purpose after the repair should go.
• Do not let the assistant assess photos with recognisable people and ask for a photo of the damage only. That is polite and saves trouble with the GDPR.
• Record which systems see the photos: the assistant, your planning package, nothing else. That goes into the processing agreements you make with the builder.

Want to see what such an assessment looks like? The chat on this site accepts a damage photo and shows right away what the assistant picks up. Photos are not stored there.