AI Damage Detection — How Findings Work

How Estimatics turns your photos into proposed damage findings: when analysis runs on each plan, what a finding contains, and how you review, correct and confirm them before they reach the estimate.

8 min read


Estimatics reads your inspection photos and turns the damage it sees into findings: structured records of what was damaged, how badly, where, and on what material. Findings are the base for everything downstream. The narrative, the estimate and the report all read from them.

The AI proposes. A person reviews and confirms. Available on every plan, including FIELD.


When analysis runs

It depends on your plan.

PlanHow photos get analyzed
SCOPE, CERTIFY and the free trialAutomatically. Each photo is analyzed shortly after it uploads, within your monthly AI allowance
FIELDWhen you ask. On the web, run it per area. On iPhone, run it per area or on photos you pick

While photos are being analyzed, the web shows "N photos being analyzed…" in Analyze.

On every plan you can also run it by hand:

  • Per area. On the web, open "Analyze" → "Areas" and click "Analyze · N left" on an area that has photos. On iPhone, open the area in "Analyze" → "AREAS" and tap "Analyze · N left". The AI reviews all of the area's photos together and consolidates the damage, surfaces and elements into a draft. The area moves to "Draft". On SCOPE and CERTIFY, use it to consolidate or re-run an area that already has proposed findings.
  • On selected photos, on iPhone. Open "Analyze" → "Analyze existing photos", pick the photos, then tap "Analyze selected photos".

There's no button to analyze a single photo on the web.

Each photo first gets a quick check for visible damage. A photo the quick check rules out isn't analyzed further and isn't counted against your allowance. In the photo viewer, a photo with no findings is labeled "Analyzed — no damage detected".

ℹ Note

Analyze needs an account that can use AI: active, or in the free trial. If you see "Could not load analysis data", check the account's billing status, then click "Retry".


What a finding contains

On insurance-claim jobs, each AI finding from a photo carries these fields:

FieldValues
Damage typeOne of 13 types: Water Damage, Wind Damage, Hail Damage, Flood Damage, Fire Damage, Smoke Damage, Mold Damage, Asbestos, Weight of Ice/Snow, Vandalism Damage, Impact Damage, Lightning Damage, or Remodel/Repairs/Renovation
SeverityLow · Moderate · High · Critical
Material affectedThe specific components, for example "drywall, baseboard, flooring"
LocationWhere in the room or structure, for example "East wall, lower 18 inches"
DescriptionTwo or three sentences of technical description
Evidence noteWhat strengthens or weakens this finding
ConfidenceHigh · Medium · Low
Photo score0–100 for the photo's quality as evidence, with a short note
Repair recommendationsTwo to four repair items visible in the photo
Policy coverageCovered · Not covered · Review required · Likely, with a note

A few of these deserve a closer look.

Severity has four levels. Critical sits above High.

Confidence is a word, not a percentage. High, Medium or Low. A number would suggest a precision the model doesn't have.

Photo score rates 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.

Policy coverage is a prompt, not a decision. Estimatics doesn't decide coverage. With no policy on file, the AI marks it Likely with a note to review it.


Where you review findings

Areas (the main view). "Analyze" opens on "Areas": one card per room or zone with its status ("Not analyzed", "Draft" or "Approved"), photos and damages. Click a card to open the area's dossier, where you review the consolidated damage, add what the photos don't show, and approve the area.

→ Areas and approval

"All findings". Every finding on the job in one list. At the top are totals ("N Findings", "AI N · Walk N · Manual N", "N Critical"). Filter by "AI Photo", "AI Walk" or "Manual". Each card shows the damage type, the photo (click to enlarge), the room (or "Assign room"), the confidence, and the blocks "Inspector narration", "AI bullets" and "Repair recommendations", with "Edit" and "Remove".

The photo viewer. Open a photo to see its AI analysis, and click "Edit finding" to correct it right there.

From Estimate, "View in Analyze" takes you straight to the finding behind a line.


Confirmed vs. proposed

Every finding is either:

  • Proposed (inferred): suggested by the AI, not yet reviewed
  • Confirmed: reviewed by a person who stands behind it

Findings you add yourself start out confirmed. When you approve an area, its proposed items are marked as confirmed ("N inferred items will be marked as confirmed."). Everything downstream knows the difference. Esti and the AI report sections see which findings are confirmed and which aren't.


The workflow

  1. Capture and assign photos to areas. Findings are much better when a photo belongs to a room. Photos without an area show up as "N photos aren't assigned to an area yet".
  2. Let the analysis run (SCOPE, CERTIFY, trial), or run it per area (any plan).
  3. Review each area. Fix what's off, remove what isn't damage, merge duplicates, and declare damage the photos don't show.
  4. Add what the AI couldn't see. Use "Add Finding" in "All findings". Manual findings count the same as AI ones, on every plan.
  5. Approve the area with "Approve area". Approved areas are what Estimate uses.
  6. Build the estimate. In Estimate, "Build with AI" builds the scope from the confirmed damage in approved areas.

→ Building the scope with AI


Adding and editing findings

Click "Add Finding" (or "Edit" on a card) and fill in:

  • "Damage Type" (required): Water Damage, Wind Damage, Hail Damage, Flood Damage, Fire Damage, Smoke Damage, Mold Damage, Asbestos, Weight of Ice/Snow, Vandalism Damage, Impact Damage, Lightning Damage, or Remodel/Repairs/Renovation
  • "Severity": Low, Moderate, High or Critical
  • "Room" (when editing), "Location", "Material Affected" and "Description"

Then click "Save Finding" or "Update Finding". You can't attach a new photo to a manual finding.


Removing findings

Click "Remove" on a card → "Delete this finding?" ("The photo will not be affected.") → "Delete".

  • The photo stays in the gallery.
  • A removed finding can't be restored from the web.
  • Deleting a photo removes its findings. If findings disappear, check whether someone deleted the photo they came from.

Reviewing well

  • Check severity against what you saw. The AI reads a photograph; you were in the room.
  • Read the evidence note. It also says what weakens the finding, and that's the sentence that matters in a dispute.
  • Remove false positives: a shadow, normal wear, a repair already made.
  • Watch the photo scores. A low-scoring photo behind a high-severity finding is the weakest link in the file.

Who can do what

Everyone on the job can see Analyze, except on Quick Quote jobs, which don't show the tab. Super Admin, Estimator and Field Inspector can run analyses, edit, remove and approve. Admin and Coordinator can only view: they see the buttons, but saving returns an error.


Usage

AI analysis draws on monthly allowances that grow with the number of people on your team and reset on the 1st of each month.

  • Area analyses: the Areas board shows "N of M analyses left this month". An analysis counts only if it produces a draft; if it fails, it isn't counted. At the limit the button reads "Monthly area-analysis quota reached".
  • Photo analyses: at the limit, new photos aren't analyzed until the allowance resets. On iPhone you'll see "Monthly analysis limit reached."

The account owner gets an email when the account is close to a limit.


What this is and isn't

AI findings support 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: October 2026 · Questions? Email support@aiestimatics.com

Last updated: October 2026 · Feedback on this article