AI-Generated Roofing Inspection Reports and Proposals

The sales rep climbs down off a roof, drives back to the office, and spends 90 minutes converting a CompanyCam photo set into a homeowner-ready inspection report and a proposal. Four days later the homeowner has a competing bid. That cycle — inspection to delivered proposal — is the single most compressible lever on a roofing P&L outside the receptionist, and estimating-from-photo is the workflow that compresses it from days to hours.

This is a transactional buyer guide for sales managers and project managers wiring AI inspection-report and supplement-drafting workflows into Roofr, AccuLynx, or JobNimbus. The pillar context lives at AI for Roofing Contractors: The 2026 Operator Playbook.

The four artifacts the rep needs to leave the kitchen table

A clean inspection-to-signed-contract motion produces four documents in one sitting: the inspection report, the aerial measurement, the tiered proposal, and (for insurance work) the draft supplement narrative. AI compresses all four.

Inspection report

The rep uploads 30–60 photos to CompanyCam at the property. The AI classifies each photo by damage type (hail bruising, wind creasing, mechanical damage, granular loss, mat exposure, flashing failure, ventilation issue) and slope (north, south, east, west), pulls in the aerial measurement, and drafts a homeowner-grade inspection narrative in 4–7 minutes. The rep reviews on the iPad in the truck, adds a sentence or two of context, and the report is ready to present.

Aerial measurement

EagleView for storm-restoration files where the carrier expects EagleView format. Hover for retail where the 3D model is part of the sales presentation. Roofr bundles both with a single click. Cycle time from order to delivered measurement is 15–60 minutes depending on the product tier, which means the measurement is ready before the rep leaves the driveway.

Tiered proposal

Good, Better, Best. AI populates each tier from your historical job-cost data and the manufacturer system selected — GAF, CertainTeed, or Owens Corning. Each tier shows the line items, the warranty length, and the financing payment (dynamic-estimating in action). The rep picks the system, the AI builds the three tiers in 90 seconds.

Draft supplement narrative

For insurance jobs. The carrier scope arrives in a PDF or Xactimate XML. The AI compares it against the Xactimate-fair scope, against the manufacturer's install instructions for the specified shingle system, and against state code, and drafts a supplement letter with each line item, code citation, and photo reference. The PM reviews, signs, and submits.

How the photo classification actually works

Modern multimodal models (Claude, GPT-5, Gemini 2.5) classify roofing damage from photos with accuracy bands of 88–94% on hail bruising, 82–90% on wind creasing, and 78–88% on granular loss in the field. The accuracy depends on photo quality — lighting, angle, distance — which is why the CompanyCam capture workflow matters. Reps who take photos within 3 feet of the damage with the sun at their back hit the high end of those bands. Reps who shoot from 20 feet in cloud cover hit the low end.

The right operator move is a one-page photo SOP that ships with every rep's iPad: 30+ photos, 3-foot proximity on damage shots, 4 slope-overview shots, 4 ventilation and flashing shots, 2 ground-level shots showing the dwelling and adjacent neighbor roofs. Adherence to the SOP lifts classification accuracy 6–9 points across the board.

The supplement draft workflow in detail

This is where the dollar moves. Most carrier first-write scopes under-scope 12–22 line items per claim. Drip edge, ice-and-water shield to code-plus-2-courses, R&R felt vs. re-lay, satellite detach-and-reset, steep-charge above 7/12, multi-story above two, granular loss treatments, and the depreciation recovery on aged shingle systems.

Claude (Anthropic) loaded with the relevant state's prescriptive code sections, the GAF or CertainTeed install instructions, and a corpus of previously approved supplement narratives from your shop produces a supplement letter that runs 4–7 pages, cites each line item with the code section and install-instruction page reference, and includes the photo evidence inline. PMs we work with report cutting their supplement-writing time from 90 minutes to 18 and lifting per-claim approved supplement value from $1,400 average to $2,600–$4,200.

The National Roofing Contractors Association (NRCA) maintains a state-by-state public-adjuster rule index that matters here — some states restrict what a roofing contractor can put in a supplement narrative without a public adjuster license. The AI workflow respects those boundaries.

Tooling stack

The configurations that consistently land:

• CompanyCam + Roofr + AccuLynx + Claude project. The most common storm-restoration stack. CompanyCam captures, Roofr measures and produces the proposal, AccuLynx is the system of record, Claude drafts the inspection narrative and supplement.
• CompanyCam + Hover + JobNimbus + Claude project. The retail-leaning version. Hover's 3D model is the proposal anchor.
• CompanyCam + EagleView direct + AccuLynx + Claude project. For shops that want the EagleView format because their dominant carriers prefer it.

All three have AI tagging and report-drafting tiers from the vendors themselves. The Claude project sits on top and handles the supplement narrative, which the vendor tooling does not do well.

The 9-day rollout

• Day 1–2: Audit. Pull 30 recent inspection reports and 30 recent supplements. Time how long each took. Identify the high-leverage edits the PM consistently makes.
• Day 3: Stand up the Claude project with state code, install instructions, and your shop's previously approved supplement library.
• Day 4–5: Pilot on three reps. Each rep runs the new workflow on every inspection for two days. PM reviews every output.
• Day 6: Tune the prompts based on the PM's edits. Most shops find 6–9 prompt refinements move the output to PM-ready by day 6.
• Day 7: Cut. All reps on the new workflow. PMs review supplements before submission.
• Day 8–9: Measure. Inspection-to-proposal cycle time, retail close rate, supplement approval rate.

ROI

For a 5-crew, 320-claim-year shop: $400K–$900K of incremental net revenue from supplement lift, $1.2M of incremental top-line from cycle-time compression on retail, payback in 8–12 weeks. Full math in Roofing AI ROI: What a Storm-Restoration Shop Actually Gains. Cross-reference the /ai-enablement operator framework for the underlying patterns.

FAQ

Q: Does the AI replace the rep's eye on the roof? No. The rep still walks the roof and confirms the damage. The AI compresses the post-walk paperwork.

Q: Will carriers reject AI-drafted supplements? No. Carriers read narratives, not authors. A well-drafted supplement narrative with proper code citations and photo evidence approves at the same rate or higher than a hand-drafted one. The 14–22 point approval lift in operator survey data from Roofers Coffee Shop operator panels and AccuLynx benchmark studies bears this out.

Q: What about Xactimate? AI doesn't replace Xactimate. Xactimate produces the line items and pricing; the AI drafts the narrative that defends those line items. Both tools stay in the workflow.

Q: How does this work for commercial TPO and EPDM jobs? Same pattern, different manufacturer corpus. The AI loads the GAF EverGuard or CertainTeed Flintlastic install instructions instead of the architectural shingle library. The supplement workflow is even higher-dollar because commercial supplements run 25–40 line items.

Q: Can the AI quote price on the spot? Yes, within the proposal generator (Roofr or AccuLynx). The pricing comes from your historical job-cost data and your current supplier pricing, not from the AI guessing.

Q: What if our rep doesn't have an iPad in the field? CompanyCam runs on iPhone and Android equally well. The Claude review happens back at the office or via a phone-based interface. The cycle-time compression is from 4 days to 4 hours either way.

Ready to compress your inspection-to-proposal cycle? Book a 30-minute operator review or read the full vertical playbook at /ai-for/roofers.