From a photo of the fault to a priced quote: AI diagnosis in practice

AI can look at a photo of a fault, name the parts it can see, suggest likely causes and draft a priced quote from your supplier prices. It can't test anything, so it can't diagnose. Use it to turn a customer's photo into a draft quote or a list of questions, then confirm the fault on site before you commit to the price.

Can AI diagnose a fault from a photo?

It can suggest a diagnosis, not confirm one. Image tools such as ChatGPT can read the label on a boiler, the rating on a breaker or the code on a display, and match what they see to common faults. That's often enough to know which part to bring.

OpenAI lists the limits of its own image input plainly: it "may generate incorrect descriptions or captions", may give approximate counts, struggles with tasks that need precise spatial location, and may misread rotated or upside-down text or images. A photo of a corroded joint looks the same whether the leak is there or 30 cm up inside the wall.

How does a photo become a priced quote?

The photo is the start of the chain. The value comes from linking what the AI sees to your supplier's prices and your labour rates, so the draft is priced the way you'd price it.

1. Customer sends a photo.

   By WhatsApp, text or email. Ask for the photos in the checklist below.

2. AI reads the photo.

   Make and model, part numbers, visible damage, error codes. It lists what it can see and what it can't.

3. AI suggests likely causes and the parts needed.

   Ranked, with the questions that would rule each one in or out.

4. Parts are priced from your supplier.

   At your trade price, not a web retail price.

5. Labour and margin are added from your rate card.

   The same way you'd price it yourself. See teaching AI your rate card.

6. You check and send.

   Either a fixed price, a price range subject to a site check, or a request for one more photo.

What does this look like for a real firm?

An electrician in East Anglia, working alone with two apprentices on solar PV, EV chargers, heat pumps and landlord work, uses this chain as part of his day. Customers send a photo of the fault. The AI identifies the kit, finds the part at his supplier's price and drafts a quote for him to check.

His admin used to happen only on Monday mornings. The same setup moved his supplier invoices into Xero automatically and checked them for price rises, so he dropped the £75 a month he was paying a bookkeeper. He also won an 11-job solar contract, and describes the time saved as "hours a week".

Notice what the draft does: it doesn't promise a fix from the photo. It prices the visit to find the fault. That is the honest answer for most electrical faults.

What photos should you ask customers for?

Ask for three or four specific shots, not "a photo of the problem". The model plate and a wide shot matter as much as the close-up of the damage.

Ask for photos straight on and the right way up: AI tools misread rotated text. For boilers, an error code photo plus the model plate usually tells you more than any close-up. See boiler fault codes for reading them.

When can you price a job from a photo alone?

When the fix is a visible, like-for-like replacement and nothing about safety depends on what you can't see. A cracked socket faceplate, a broken radiator valve head, a missing ridge tile on a bungalow. Anything that needs testing, opening up or a safety judgement needs a visit.

Electrical work has its own rule. In England, notifiable work in homes, such as a new circuit or work in a bathroom, must be self-certified by a competent person scheme member or checked by building control, as Approved Document P sets out. Scotland and Northern Ireland have their own building regulations.

What goes wrong with AI fault diagnosis?

The common failures are confident wrong answers and quotes that turn into promises. An AI that sees a corroded valve will happily blame the valve. If you send that as a fixed price, the customer will hold you to it when the real fault is the pipe behind it.

Three habits avoid most of it. Always say in the quote what it assumes. Price the visit, not the fix, whenever testing is needed. Keep the photo with the quote so you can show what you priced from. For more on catching AI mistakes, read when AI gets it wrong.

Common questions