How it works

No black box. Here's exactly what we score.

A top-down satellite shot and a ground-level Street View of the façade are read together into one assessment — not two conflicting opinions. Then the AI assigns its own 0–100 buy-probability score. No hidden formula. Every score comes with a plain-English verdict and the exact reasons behind the number, click the card, see what the AI saw.

Two inputs

The roof tells us what's there. You tell us what you want.

Input 1 · Imagery

What the AI reads — top-down AND street-level

  • Satellite (top-down) — material, footprint, complexity, facets, vents, solar, tree overhang
  • Street View (ground-level) — the façade the AI also reads for side-visible wear
  • Granule loss, algae streaking, curling shingles, patched sections, rusted flashing — what a top-down tile can't see
  • Visible condition — moss, missing shingles, dark patches, tarps, partial repairs
  • Roof age — read off the actual shingles, not just the building's year
  • Property records — building/renovation year, value, owner-occupied vs absentee
  • The immediate neighbours — read from a wider view centred on the target, so 'the two either side went new and this one didn't' is a signal we can actually attribute
  • Owner equity — estimated from value against what's still owed, because equity decides whether someone CAN say yes
  • Both views read together into ONE assessment — not two conflicting opinions — with the street read weighted higher when the roof's clearly visible

Input 2 · Your business

What we learn from your website

  • Materials you install — residential asphalt? metal? tile? commercial?
  • Sales angle — insurance retrofit, age-driven, storm response, premium upgrade
  • Geographic focus — service area, urban vs suburban patterns
  • Brand voice — pulled into the door-hanger PDF copy automatically
  • Logo + brand color — pulled for the PDF and dashboard
  • Contact phone — for the door-hanger CTA
  • One-paragraph business summary — used to judge fit per roof

Same map, different roofers. An asphalt shop and a commercial flat specialist get completely different lead lists from the same polygon. The roof's score depends on your profile.

Sample score

Every score is broken down.

Click any lead in the dashboard to see exactly what the AI saw. Here's what that looks like for one home.

Lead · 1234 Ashbury Ln, Plano TX

Aging asphalt, replacement-ready

87

High likelihood

The AI's verdict

"Decent knock — mature asphalt with recent hail nearby and three neighbors already re-roofed. Worth a conversation."

Signals

Asphalt shingle (matches your install profile)Visible granule lossEstimated 18–22 yrs3 neighbors re-roofed within 200ftModerate moss on north slopeNo tarps, no recent repair patchesSingle-story, simple gable

AI-generated door pitch

"Hi, I noticed three homes on Ashbury were replaced this year. Your roof is about the same age and we'd love to show you what yours might look like, free, just 10 minutes."

Why this number

What raised it

  • Granule loss visible on the street read
  • Roof age estimated 18–22 yrs — past its prime
  • 3 neighbors re-roofed within 200 ft
  • Hail within your storm window

What lowered it

  • Simple single-story gable — smaller job
  • No tarps or active-leak signs

One-click unlock

A scored roof is half the job. Then you reach the owner.

One click on any lead unlocks the full owner and property picture, no separate skip-tracing tool, no bought list.

Two axes

Why one score was never enough

  • NEED — how badly the roof needs replacing. Condition and roof age, read off the shingles
  • NOW — whether anything is forcing a decision on THIS house: a tarp or missing shingles, hail damage on this roof, a purchase in the last 18 months, patching already attempted, neighbours who have re-roofed
  • A one-line 'why now' in plain words, or an honest 'no trigger yet'
  • Neighbourhood hail does NOT raise NOW — every home in the scan shares it, so it cannot separate one door from the next. It belongs in your pitch, not the ranking
  • Where hail is on record: roughly how long the owner has left to file, from the typical notice deadline in that state. An estimate — verify with the carrier
  • Blending the two is how every other list ranks a tired roof nobody's thinking about above a homeowner already ringing round

Owner contact

Who to call, and who's safe to call

  • Owner name(s) and mailing address
  • Phone numbers + emails — each phone flagged clear vs on the DNC list
  • Owner-occupied vs absentee / out-of-area owner
  • Calls are manual and DNC-scrubbed only (TCPA) — no texts, no robocalls

Property intel

The whole property record, sized before you knock

  • Estimated (AVM) market value + a confidence score
  • Year built, last sale date + price
  • Beds, baths, living square footage
  • Lot size, stories, garage
  • Annual property tax and mortgage lender

An unlock only counts against your allowance if we actually find a phone or email. A miss never costs you a credit.

Business-fit examples

Same map. Different leads.

Three roofers running Roofbird on the exact same zip would see three very different lead lists. Here's why.

Residential asphalt shop

Promoted

Aging asphalt, hail-touched neighborhoods, 18+ year roofs

Demoted

Tile, slate, metal, commercial flat → auto-low

Commercial flat specialist

Promoted

TPO, EPDM, PVC membranes, ponding signs, parapet wear

Demoted

All residential pitched → auto-low

Tile + slate restorer

Promoted

Tile roofs with visible cracks, slate with missing pieces

Demoted

Asphalt, metal → auto-low (wrong material)

The keep-filter

Roofs the AI rates low don't count against your quota.

Every roof in your polygon gets scored. But only high and medium likelihood roofs land in your list. Low-likelihood roofs (recently replaced, wrong material, no visible wear) get filtered out and don't burn quota.

01

In polygon

3,200 homes

02

AI scores all

3,200

03

Low filtered out

2,750 (no charge)

04

Kept on your list

450

You only pay quota for leads worth knocking. The 2,750 filtered roofs are still scored and stored, click Get leads next month and the AI moves on to roofs it hasn't shown you yet. Every scan is logged, so you can filter your leads down to a single scan any time.

Honest limits

What we don't claim.

We do NOT detect yesterday's storm damage from satellite.

Public satellite tiles are months old. Storm-chasers should use Roofbird for baseline condition + neighborhood signals, and pair with NOAA storm overlays we surface separately.

We do NOT have a street-level view of every home.

The ground-level read uses existing Street View — it's blocked by trees on some homes and missing on newer or private streets. When it's not usable, we fall back to the satellite read and say so. It's also a periodic capture, not live.

We do NOT guarantee every score is right.

The AI gets surprised by tarps, weird shadows, and freshly-painted-but-old asphalt. That's why every score is explainable, you decide whether to trust it.

We do NOT replace door-knocking instincts.

A great rep with a Roofbird list will out-close a mediocre rep every time. We sort the doors. You still have to knock.

See it on real homes

Score a real polygon in 60 seconds.

10 free leads, no card. Drop your website so the AI learns your business, sketch your area, click Get leads.