AI Damage Detection — How Findings Work

Estimatics reads your inspection photos and turns each piece of damage into a finding — a structured record of what was damaged, how badly, where, and on what material. Findings are the spine of everything downstream: the narrative, the scope, and the report all read from them.

Available on every plan, including FIELD.

---

You trigger it. It does not run on its own.

This is the most important thing on the page, and the most commonly misunderstood.

Analysis does not start by itself when photos finish uploading. You run it — per photo, or per area with the Analyze button on the area dossier. Nothing is sent to the AI, and nothing is billed against your usage, until you ask for it.

That's deliberate. Automatic analysis of every photo the moment it lands would spend your allowance on blurry frames, duplicates, and the twelve shots of the same wall you took to be safe.

---

What a finding contains

Every finding carries these fields:

Two of those deserve a closer look.

Severity has four levels, not three. critical sits above high. If you've been reading a three-level scale somewhere, this is the real one.

Confidence is qualitative, not a percentage. It's high, medium, or low. There is no 87% — a number would imply a precision the model doesn't have.

Photo score is a 0–100 rating of the photograph, not the damage: clarity out of 40, lighting out of 30, damage visibility out of 30. A low score is a prompt to retake the shot while you're still on site.

---

Confirmed vs. inferred

Alongside those fields, every finding carries a status:

• Inferred — proposed by the AI, not yet reviewed by you
• Confirmed — you've reviewed it and stood behind it

Everything downstream respects that distinction. It's also your review queue: work the inferred, low-confidence, high-value findings first.

---

The workflow

1. Capture and assign to areas. Findings are far better when a photo belongs to a room — the AI knows what it's looking at.
2. Run the analysis — on a photo, or on a whole area from its dossier.
3. Review in Analyze → Findings. Confirm what's right, correct what's off, delete what isn't damage.
4. Add what the AI couldn't see. Manual findings sit in the same list — on every plan, not a special tier.
5. Flag what belongs in the scope with +Scope. That's the queue AI Scope Generation reads from.

---

Reviewing well

• Check severity against what you saw. The AI reads a photograph; you were in the room.
• Read the evidence note. It's written to tell you what weakens the finding too — that's the sentence that matters in a dispute.
• Treat policy coverage as a prompt, not a determination. Estimatics doesn't adjudicate coverage. With no policy on file, the note says so.
• Delete false positives. A shadow, a stain that's normal wear, a repair already made. Deleted findings are soft-deleted, so nothing is lost.
• Watch the photo scores. A low-scoring photo behind a high-severity finding is the weakest link in the file.

---

Usage

AI analysis draws on a monthly allowance that scales with your team. The area dossier shows how many area analyses you have left, and warns you before you run out. Photo analysis and Copilot usage are metered the same way.

---

What this is and isn't

AI findings supplement professional judgment; they don't replace it. Estimatics doesn't decide coverage, assign liability, or certify that damage occurred. Review every finding before it reaches a deliverable — the assessment that goes out is yours.

---

---

Last updated: September 2026 · Questions? Email support@aiestimatics.com