"AI-powered" is doing a lot of work in a lot of marketing right now, so here is what it means in practice on an AyeInspect survey — and what it deliberately does not mean.
What happens to a photograph
Each aerial image is passed to a vision model with a structured brief: find defects, place a box around each one, name the category, describe it, grade the severity and state how confident you are. The model returns coordinates as proportions of the image, which is how the circles land in the right place whether you view the report on a phone or a 27-inch monitor.
What it is genuinely good at
- Consistency — it looks at the eightieth photograph of the day exactly as carefully as the first.
- Small repeated patterns: hail bruising, granule loss, hairline cracking across a run of tiles.
- Never skipping an image because the roof "looked fine" in the previous one.
- Drafting clear, non-jargon descriptions a homeowner can read without a glossary.
Where it gets things wrong
Shadows get mistaken for gaps. Wet patches after rain look like staining. A repaired tile in a slightly different shade can be flagged as damage. Reflections off rooflights confuse it. These are exactly the errors a person with roofing knowledge spots in seconds, which is why the workflow is built the way it is.
Every finding is reviewed by a human
Nothing reaches a client automatically. Each detection arrives with a confidence score into a review workspace where it is approved, edited or rejected. A surveyor can also draw a circle around anything the model missed and write the description themselves. The report a client opens contains only findings a person has signed off.
Why bother with AI at all, then?
Because it changes the economics. Reviewing pre-annotated photographs takes a fraction of the time of examining raw images, which is what makes a genuinely thorough survey viable at £99 rather than several hundred. The judgement stays human; the donkey work does not.