Incidents

AI photo damage assessment: how it works and what to expect

What an AI can and can't see in a photo of a breakdown, how it helps prioritise incidents and why it doesn't replace the loss adjuster.

María Velasco · 10 Sept 2026 · 4 min read

1 phototo prioritise a breakdown

A photo of a stain on the ceiling says far more than it seems: whether the ring is fresh or old, whether there's active dripping, whether it's close to an electrical installation. An experienced account manager reads it in three seconds. The question is whether an AI can do that first reading just as fast, at any hour and with the same criteria across every community.

The short answer: it can do the first reading, not the loss assessment. And that difference is exactly what makes it useful.

What the AI does when it "looks" at a photo

Multimodal AI models analyse an image together with the text that comes with it. Applied to an incident, they can:

Identify the type of damage: damp, leak, crack, broken glass, detachment, lighting fault.
Locate the affected element: ceiling, façade, downpipe, electrical panel, garage door.
Spot signs of severity: visible water, proximity to electrical installations, fallen pieces, size of the affected area.
Cross-check it with the resident's description: "it's been dripping since last night" weighs differently from "I noticed it today".
Suggest a category and urgency level, using the same criteria for every incident.

What the AI can't see

Being honest about the limits is what prevents nasty surprises:

It can't measure what isn't visible. Damp inside a wall or a leak in a concealed pipe doesn't show up in the photo.
It depends on the image. A dark, blurry or too-close photo reduces reliability.
It doesn't determine liability. Whether the source is private or communal is for a technician to decide after an inspection.
It doesn't replace the insurer's loss adjuster, nor does it produce a technically valid report.

That's why the right use is triage: quickly deciding what needs immediate attention, which specialist to send and with what prior information.

How it helps a property management firm day to day

Use	What it brings
Prioritising	Separating an active leak from an old stain without an account manager reviewing every photo
Sending the right specialist	The plumber doesn't turn up to an electrical problem
Briefing the contractor	They arrive knowing what to expect and with the right materials
Documenting for the insurer	Photo with date, time and description from minute one
Community history	Seeing whether the same spot keeps recurring and making the case for works at the owners' meeting

How to get residents to send a good photo

The quality of the analysis starts on the resident's phone. Share these four tips on the noticeboard or in the welcome message of your incident channel:

One wide shot to show where it is and one close-up to show the detail
Good lighting: switch on the landing light or use the flash
If there's water, show whether it's dripping or still (a 5-second video helps)
Never get close to cables, gas or anything that could fall

Where Wabee is today and where it's heading

Today, in Wabee, owners report incidents with a photo from the app or via WhatsApp, and the AI classifies urgency and alerts the contractor without the property manager stepping in.

The next step we're working on is damage assessment from the image: estimating extent and severity, suggesting the type of intervention and producing a documented pre-report for the insurer and the owners' meeting. Always as support for the judgement of the property manager and the technician, never as a replacement.

Frequently asked questions

Does a photo assessment count as a loss assessment for the insurer? No. It serves as initial documentation of the claim. The loss assessment is carried out by the professional appointed by the insurer.

What if a resident sends a photo that doesn't make sense? A good system asks for another image or more details before classifying, just as an account manager would.

What about the privacy of photos taken inside someone's home? They must be treated as personal data: role-based access and hosting with proper safeguards. At Wabee, data is hosted in the EU and is never used to train AI models.

Can the AI get the classification wrong? Yes, like any person. That's why the criteria must be reviewable and the property manager must be able to correct the classification at any time.

Want to see how a photo becomes a classified, routed incident in seconds? Book a Wabee demo