RoofGenius.ai

Hail and storm damage, documented photo by photo.

Upload the photos you already take. Every image is classified for hail, wind, granule loss and flashing damage, then delivered as one insurance-ready report.

Book a demo

Phone, drone, or aerial photos. Findings scored, never assumed.

0
Damage classes identified
0+
Confidence floor per finding
~0s
Typical job turnaround
Why it matters

The damage you can't prove is recovery you don't get.

Most claims aren't lost on price — they're lost on evidence. Damage detection puts a classified, scored, written-up finding behind every line before the supplement is even started.

"No visible damage" ends the claim

A folder of untitled photos is not evidence. Classified findings with severity and confidence on each image make the damage hard to wave off in the first review.

Manual photo review burns the estimator's day

Nobody should be squinting at 80 phone shots and typing captions. Each photo is read and written up automatically in about the time it takes to make coffee.

Undocumented hits become unbillable scope

Findings carry their component and unit of measure, so detected damage flows into scope lines instead of being retyped — or forgotten.

Inside the report

What the AI returns for every single photo.

Not a verdict on the roof — a structured record per image. Each damage instance the model can clearly see becomes its own row, and anything it isn't sure about is dropped before it reaches you.

Damage class
One of 17 mapped roof components
Severity
Minor · Moderate · Severe
Confidence
50–100, sub-50 discarded
Description
Written per photo, report-ready
Location
Where in the frame the hit sits
Unit
SQ · LF · EA for scope lines

Damage classes it identifies

Mapped to roof component and unit of measure.

Shingle
Hail impact / bruisingGranule lossCracked shingleMissing shingleLifted shingleCreased shingleGeneral impact
Soft metals
DentsVent damageGutter damage
System
Damaged flashingExposed underlaymentWater damage / staining
Tile
Broken tileCracked tile
The checklist

What gets checked on every photo.

This is the roof inspection checklist the model works from. Each class maps to a roof component and the unit that carries into a scope line — so the finding is already estimator-ready when you open it.

Damage classComponentUnit
Hail impact / bruisingField shingleSQ
Granule lossField shingleSQ
Cracked shingleField shingleSQ
Missing shingleField shingleSQ
Lifted shingleField shingleSQ
Creased shingleField shingleSQ
General impactRoof surfaceSQ
DentsSoft metalsEA
Vent damageRoof ventEA
Gutter damageGutter / downspoutLF
Damaged flashingFlashingLF
Exposed underlaymentUnderlaymentSQ
Water damage / stainingDecking / interiorSQ
Broken tileTile fieldEA
Cracked tileTile fieldEA
Wind damageRoof surface / rakeSQ
Ridge / hip damageRidge capLF
The deliverable

What a finished roof inspection report contains.

Six sections, in this order. No template to fill in — every part is written from the photos and the property record.

01

Cover & property record

Address, inspection date, date of loss, and the elevations covered — so the reviewer knows exactly what was looked at before they read a single finding.

02

Job-level damage summary

Overall severity, the classes found, and how many photos support each one. Derived from the findings, not typed by hand.

03

Photo-by-photo evidence

Every image with its detected damage, severity, confidence score, and a written inspection note. This is the section that ends the "no visible damage" argument.

04

Verified storm history

NOAA hail and wind events matched to the property coordinates and date of loss, with distance and reported stone size. If nothing qualifies, the report says so.

05

Building code support

State and local code requirements attached to the findings that trigger them, so the repair is argued from the code book instead of asserted.

06

Measurements & scope

The measurement report and the scope lines built from the findings live on the same file — one document to the adjuster, not three.

All sample documents
Cause of loss

Hail, wind, mechanical, or wear.

Most denials are cause-of-loss arguments, not damage arguments. Findings are classified on what is visible in the frame, then checked against verified storm events for that address and date.

Hail

Looks like: Random-pattern impacts, granule loss, soft bruised mat

Confirmed by: Round dents on gutters, vents, drip edge in the same direction

Wind

Looks like: Creased tabs, lifted shingles, missing shingles

Confirmed by: Concentrated on exposed elevations and along the rake

Mechanical

Looks like: Scuffs, broken tabs, crushed granules in a line

Confirmed by: Clusters on traffic paths and around HVAC or satellite mounts

Wear / defect

Looks like: Blistering, curling, uniform granule thinning

Confirmed by: Even across every slope and unrelated to any storm date

How it works

Photos in, adjuster-ready evidence out.

01

Upload

Phone photos, drone captures, aerial imagery, or a synced photo album. Any angle, any resolution, no capture protocol to memorize.

02

Pre-check

Every image is validated as a roof or property photo first, so selfies, receipts, and driveway shots never contaminate the findings.

03

Classify

A vision model reads each photo individually and returns every clearly visible damage instance with class, severity, confidence, and frame location.

04

Support

Findings are paired with written photo notes, state and local building-code research, and verified storm history for the date of loss.

05

Deliver

Everything merges into one job file — damage report, measurements, estimate, and supplement in a single document for the adjuster.

Built for storm work

Engineered to win re-inspections.

Per-photo classification

Each image is analyzed on its own and merged into one job-level summary — not a single guess made from a thumbnail grid.

Conservative by design

The model is instructed that false positives are worse than misses, and anything under 50 confidence never reaches the report.

Findings that carry units

Every damage class maps to a component and unit of measure, so evidence turns into scope lines instead of retyping.

Code-backed reasoning

State and local building-code research attaches to the findings that need it, so the repair requirement is argued, not asserted.

Storm history on the file

Verified hail and wind events for the property location and date of loss sit alongside the photo evidence.

Automatic photo write-ups

Professional descriptions and inspection notes are generated per image, so the report reads like a field inspection.

Minute-scale turnaround

Photos are analyzed in parallel. No queues, no offshore reviewers, no next-day delivery window.

One connected job file

Damage findings, measurements, estimate, and supplement live on the same file and ship as one document.

FAQ

Questions, answered.

How do you inspect a roof for hail damage?+

Walk the slopes you can safely reach and photograph each elevation, the ridge, valleys, penetrations and every soft metal — gutters, downspouts, vents, drip edge, flashing. Hail leaves random-pattern impacts with granule loss and a soft, bruised mat; soft metals show round dents that confirm stone size and direction. Upload the photos and each image is classified individually: damage class, severity, confidence score, and a written note. You are documenting evidence, not just looking at a roof.

What should be included in a roof inspection report?+

Property and date of loss, a photo-by-photo record with a written description of each image, the damage class and severity for every finding, a confidence score so the reviewer can see what is certain and what is borderline, the affected roof component with its unit of measure, verified storm history for that address, and the building-code requirements that drive the repair. Anything less and the adjuster is guessing.

What does a roof inspection report look like?+

A cover with the address and inspection date, a job-level damage summary, then one section per photo — the image, the detected damage, severity, confidence, and the inspection note written for it. Storm history and code citations attach at the back, and the measurement report and scope live on the same file, so the adjuster gets one document instead of three unrelated PDFs.

How do you tell hail damage from wind, mechanical, or manufacturing defect?+

Hail is random-pattern with granule loss and dented soft metals in the same directional pattern. Wind shows creasing, lifted tabs, and missing shingles concentrated on the exposed elevations and along the rake. Mechanical damage clusters on traffic paths and around HVAC. Blistering and defect are uniform across the slope and do not match a storm date. The model classifies each hit on what it can actually see in the frame, and the verified storm history on the file either supports the loss date or it does not.

What kinds of photos work — phone, drone, or aerial imagery?+

All of them. iPhone or Android photos taken on the roof, DJI drone captures, aerial imagery, or a synced CompanyCam album. Photos are normalized before analysis, so mixed resolutions, oblique angles, and imperfect light are fine. There is no special capture protocol to learn.

How accurate is it, and does it invent damage?+

The model is deliberately conservative: it is instructed that false positives are worse than missed items, and it must return a confidence score for every finding. Anything under 50 confidence is discarded before you ever see it. Job-level severity is derived from the highest-confidence finding, so borderline calls are visible rather than buried — your estimator confirms or removes them in review. We do not publish an accuracy percentage, because a single number across every roof type and lighting condition would be marketing, not measurement.

Does each photo get written up, or just tagged?+

Both. Beyond the damage tags, the system generates a professional photo description and inspection note for each image, so the report reads like a field write-up instead of a folder of untitled JPEGs.

Is the storm history verified, or just a date the homeowner gave you?+

Verified. Hail and wind events are pulled from NOAA storm-event data for the property coordinates and matched against the date of loss, with distance and reported stone size. If no qualifying event exists near that address, the report says so — that is worth knowing before you file, not after.

Do adjusters accept these reports?+

The report format is what a reviewer expects from a competent field inspection: photo-by-photo evidence, damage classification, severity, confidence, and a written description per image, with code and storm support attached. It removes the ambiguity that drives "no visible damage" denials and re-inspection requests.

How does it connect to supplements and measurements?+

Damage findings auto-attach to the same job file as your measurement report, estimate, and supplement. Detected components carry their unit of measure, so the scope lines are built from the evidence instead of retyped from it — one connected file to the adjuster instead of three unrelated PDFs.

How fast is the turnaround?+

Analysis runs per photo in parallel, so a typical 40–80 photo job finishes in about a minute. No queue, no offshore review team, no waiting on a human reviewer.

How do I get started?+

Subscribe to the $149/month Retail plan to use damage detection, measurements, proposals, and job tracking across every job.

Catch the damage adjusters miss.

Classify, document, and deliver evidence-grade reports on every storm job.

Book a demo