Hail Damage Assessment From Satellite: What AI Can (and Can't) See

A common question I get from roofers evaluating AI tools: "can satellite imagery actually detect hail damage, or is it just marketing?" The honest answer: yes, but with limits. Satellite + AI can detect a meaningful subset of hail damage signals at scale. It can't replace in-person inspection — and any vendor claiming otherwise is overselling.

This post is about exactly what AI can spot from above, exactly what it can't, and how to use the combination intelligently.

Why hail damage is hard to assess from above

Hail damage on roofs splits into two categories:

Cosmetic / surface damage:

• Granule displacement (lighter spots where hail knocked granules off)
• Visible bruising on certain shingle types
• Missing tabs (severe impacts only)
• Tarp installation (homeowner's emergency repair)

Structural / sub-surface damage:

• Mat damage under intact-looking shingles
• Cracked decking
• Compromised underlayment
• Bruise depth (the actual impact severity)

The cosmetic signals are visible from above. The structural signals are NOT — they require ground-truth inspection where the roofer touches the shingle, presses for soft spots, and inspects the underside.

This is the structural reason why satellite AI for hail damage is a screening tool, not a diagnostic tool.

Signals AI CAN pick up reliably

Internal validation data from our network (~200 ground-truth comparisons) on what modern vision AI detects:

Signal	Accuracy	Notes
Visible tarp	99%	Bright blue is unmistakable from satellite
Missing tab sections	96%	High contrast vs intact roof
Granule displacement (severity)	88%	Visible color/texture variation
Tree-fall debris on roof	92%	Distinct shape signatures
Circular impact bruising	78%	Better on architectural asphalt than 3-tab
Ridge cap damage	84%	Often visible from oblique satellite angle
Visible work-in-progress	95%	Dumpsters, debris, crews are unambiguous

The "visible work in progress" signal is the most underrated — when an AI tool spots an active job, you've got 14-28 days before another roofer signs the contract. You can knock the neighbors NOW for the cascade effect.

Signals AI CAN'T pick up

Signal	Why AI misses it	Workaround
Bruise depth	Bruises visible from above but depth requires touch	Ground inspection
Sub-tab mat damage	Hidden by intact top layer	Ground inspection
Soft decking	Not visible at all	Walk-through inspection
Hidden leaks	Interior damage, not roof surface	Homeowner conversation
Recent (under 7 days) damage	Satellite imagery lags	Use NOAA event data as proxy
Damage covered by panels/solar	Obstruction	Ground inspection
Wind damage to flashings	Detail too small for satellite	Ground inspection

The pattern: AI sees surface damage well. AI misses sub-surface damage entirely. This is why "AI replaces inspection" is wrong — the AI screens which homes have damage worth inspecting, but the inspection itself remains essential.

The screen-and-verify workflow

The right way to use satellite hail damage assessment:

Step 1: AI screening (5 min per zip)

After a hail event, run AI scan on the impacted swath. AI flags ~10-20% of properties with visible damage signatures.

Step 2: Triage (10 min)

Review the AI flags. Drop obvious false positives:

• Commercial flat roofs (most consumer AI tools score these wrong)
• Multi-unit complexes (apartment buildings often flag as residential)
• Recently-replaced roofs the imagery hasn't caught up to

You're left with ~5-15% of the swath as real candidates.

Step 3: Door-knock + inspect (3-4 hrs)

For each candidate, knock + offer free inspection. The conversion rate to inspection-scheduled is 30-50% with a proper opener.

Step 4: Verify on ground

In-person inspection confirms:

• Bruise depth (touch test)
• Mat damage (lift a tab)
• Decking condition (walk the roof)
• Any sub-surface signal AI missed

The screen-and-verify workflow gets you to actual inspectable prospects 5-10x faster than blanket canvassing.

Cost comparison: satellite screening vs. ground inspection

For a typical hail-belt zip with 4,000 homes:

Pure ground canvassing:

• 4 hours per 30 homes inspected
• 4,000 homes ÷ 30 = ~133 inspection days = ~$13,300 in rep labor
• Cost per actually-damaged home found: $150-300

Satellite screening + ground verification:

• AI scan: 5 min, ~$2-10 amortized subscription cost
• AI flags 400-600 candidates
• Ground inspection on top 200: 27 days = ~$2,700 in rep labor
• Cost per actually-damaged home found: $13-25

The satellite screening cuts per-damaged-home discovery cost by ~10-20x. The roofer who knocks 200 AI-flagged doors closes more jobs than the roofer who knocks 1,000 random doors.

Roofbird's DFW sample dashboard shows what AI-flagged hail damage looks like in practice. The May 9 Mesquite event surfaced 8 properties with blue-tarp signatures — one (2011 Rayburn Ave) you can verify yourself in Google Maps.

What AI evidence proves (and doesn't) for insurance

Insurance carriers vary in how they treat satellite-derived damage evidence:

What AI evidence supports:

• Pre-storm baseline (the roof looked X way before the event)
• Post-storm change (visible damage that wasn't present before)
• Property-level proof a storm hit (vs. just county-level)
• Contextual signals (tarps, neighbor replacement cascades)

What AI evidence doesn't replace:

• Touch inspection (required for claim approval in most carrier protocols)
• Adjuster verification
• Repair-vs-replace determination

The strongest insurance-claim documentation packages combine:

• AI-detected pre/post comparison
• Ground-truth photos from the roofer's inspection
• Storm-event verification from NOAA
• Permit history showing no recent replacement

When the four are stacked, claim approval rates are dramatically higher than ground inspection alone.

The future: drone integration

Satellite resolution at 5-15cm/pixel is enough to spot tarps and missing tabs. It's not enough to spot bruise patterns reliably.

The next 12-24 months will see AI roofing tools layer DRONE imagery on top of satellite — automatically triggering a drone flight for properties flagged as high-priority. Drones get 1-2cm resolution and can capture oblique angles satellite can't.

For most residential shops in 2026, drone integration is overkill. Satellite + ground inspection is sufficient. For shops doing commercial flat-roof or high-value-luxury work, drones become worth it sooner.

Working AI hail damage assessment into your workflow

The integration that works for most residential storm-chase shops:

1. Set up NOAA email alerts for your service area
2. Subscribe to one AI satellite tool with hail-damage detection (Roofbird or similar)
3. When a NOAA alert fires, run AI scan within 48-72 hours (allow time for imagery to refresh)
4. Triage AI output — drop commercial false positives, prioritize residential homes with high-confidence damage signals
5. Door-knock the top 50-100 with the storm-event opener
6. Inspect on-site — verify what AI flagged + check for sub-surface damage AI couldn't see
7. Document with both AI evidence + ground photos for insurance claims

What to do this week (even without a recent storm)

Even before the next hail event:

1. Test AI satellite damage detection on the most recent historical event in your area
2. Calibrate your eye for the visual signals (what does AI mean by "missing tabs"? "granule displacement"?)
3. Build a triage checklist for your team — what to drop as false positives, what to escalate
4. Set up the NOAA alerts so you're notified within hours of the next event

When the next significant hail event hits, your reaction speed is the competitive advantage. Shops that have the workflow pre-built capture the work; shops that scramble lose to the prepared ones.

— Jake