AI Lead Scoring for Roofing, Explained Honestly

The two questions every scoring model is really answering

Strip the marketing off any lead scoring software and you are left with two questions. How bad is this roof? And is anything forcing a decision on this house right now?

Those are different questions and they deserve different numbers. A roof can be fifteen years old with granule loss everywhere and nobody inside is thinking about it. A roof can be nine years old with a tarp on the back slope and a claim window closing. If your scoring tool collapses both into one 0-100 score, you have a number that cannot tell you which door to knock first.

So here is what the model actually looks at, why a comparative rank beats an absolute score, what confidence means when a machine is judging a photograph, and how to test any vendor's score against streets you already know.

1. What the model reads on a single roof

A roof read from imagery produces a specific, checkable list. Not a vibe. Here is the inventory:

• Roof age, read off the shingles. Not the building's year built. A 1972 ranch with a 2019 re-roof is a midlife roof, and any model that scores it as fifty years old is scoring the wrong thing.
• Wear signs. Granule loss, algae streaking, curling shingles, patched sections, rusted flashing, exposed decking.
• Active damage. A tarp, missing shingles, a board over a hole. This is the difference between a roof that is old and a roof that is failing.
• Geometry. Squares, number of planes, pitch, stories, dormers, chimneys, skylights, solar panels, a detached garage.
• Obstructions. Tree canopy over the roof, which matters because it hides the roof from the camera and because it is a maintenance problem the homeowner already knows about.

The overhead tile alone catches maybe two-thirds of that. A top-down view is good at geometry and terrible at algae streaking down a north-facing slope, because from directly above, streaking is invisible. That is why the honest version of this reads the overhead tile and a ground-level street view of the façade into one combined assessment. How Roofbird scores a roof from imagery lays out the mechanics if you want the full pipeline.

2. Why a comparative rank beats an absolute 0-100 score

This is the part most vendors hand-wave, so let's be concrete.

An absolute score is a number with no reference point. A roof that scores 72 out of 100. Seventy-two relative to what? Relative to every roof in America? Relative to the model's training set? Relative to the other roofs the vendor happened to read last Tuesday?

The practical problem is that absolute scores cluster. Run a scoring model across a normal subdivision and you get a bell curve squeezed into a narrow band. Everything is between 55 and 75. The worst roof on the block scores 68 and the best scores 74, and you are supposed to knock the 68 first. Good luck holding that in your head at a door, and good luck explaining it to a homeowner.

Worse, an absolute score is not reproducible at the doorstep. You cannot stand on the porch and say "this roof scores 72." The homeowner will ask 72 out of what, and you have nothing.

A comparative rank fixes both problems. Each roof is placed against the roofs either side of it and sorted into one of four buckets: replace-now, worn, midlife, or already done. That is a statement you can repeat at a door. "Both houses either side of you have been re-roofed in the last four years. Yours hasn't. That's why I'm here." That sentence closes. A 72 does not.

The comparison is also where the strongest social proof in the trade lives. A street where the neighbours have already gone new is the single best reason a homeowner will finally pick up the phone, and it is invisible to any model that scores houses one at a time in isolation. If you want to see the difference on a real block, a live scored area you can look at without signing up shows the ranking in place rather than described.

3. NEED and NOW are two numbers, not one

Roofbird carries two scores per lead and refuses to merge them.

NEED is how badly the roof needs replacing. Condition and roof age only. It does not care about the owner, the market, or the weather.

NOW is whether anything is forcing a decision on this specific house: visible active damage, hail damage on this roof, a purchase in the last eighteen months, patching already attempted, or neighbours who have re-roofed.

Here is the discipline that makes NOW trustworthy. Neighbourhood hail does not raise the NOW score. If a storm dropped two-inch hail across a ZIP code, every home in your scan shares that fact, which means it cannot separate one door from the next. It belongs in your pitch, not in the ranking. Storm-chasing lists do the opposite: they rank by proximity to a hail swath and hand you a list where every house looks equally urgent, which is another way of saying the list is unranked.

Most homes in a settled neighbourhood honestly have no trigger. A tool that says so is more useful than one that invents urgency to look busy. Where a hail event is on record, Roofbird shows roughly how long the homeowner has left to file a claim based on the typical notice deadline in that state. Treat it as a countdown to start the conversation, not a legal deadline. Verify with the carrier. The US hail map is where the event history lives.

4. What "confidence" actually means

Confidence in roof scoring is not a measure of how sure the model is about the roof. It is a measure of how much of the roof the model could see.

This distinction matters more than anything else in this article. A contractor in Missoula drove nine doors off his own list and eight of them had already been re-roofed. His neighbourhood had been ranked from a July 2016 aerial photograph while a 2025 street-level shot of the same houses sat unread. The roofs had not changed since we read them. The picture had.

Street-level imagery is newer than the aerial on roughly 85% of roofs, by about five years on average. So every shortlist gets checked against the newer picture before you see it, anything already done moves down and gets said out loud, and roofs that genuinely cannot be judged, behind tree canopy or set too far back from the road, are labelled as exactly that instead of mixed in with the confident ones.

The test that convinced me: where the street camera could see the roofs, the model matched what the contractor found on the ground four out of four. Where it could not, zero out of five. That is what a confidence label is for. It is not decoration on a score. It is the difference between a list you can drive and a list that wastes a Tuesday.

Where the aerial and the street photo disagree, the newer one wins and both capture dates are shown. A roof whose own description says it was recently replaced never reaches the top of a list.

5. How to sanity-check any vendor on your own streets

Do not take a demo at face value. Run this test instead, and run it on a block you have personally walked.

1. Pick ten houses you know. Include two you know were re-roofed recently and two you know are original.
2. Ask for the imagery date on each. If the vendor cannot tell you when the photograph was taken, the score is unfalsifiable and you should stop there.
3. Check the sort order. The two you know were re-roofed should not be at the top. If they are, the model is reading old imagery.
4. Look for confidence labels. A vendor that presents every roof as equally knowable is hiding the ones it cannot see.
5. Ask what the score is relative to. If the answer is a national scale or a training set, you have an absolute score and it will cluster.
6. Check whether neighbourhood hail moves the ranking. If it does, the list is not ranked, it is filtered by storm.
7. Read the property record on a house you own or have sold. Year built, last sale date and price, beds and baths. This is public data and it is either right or it is not.

If a vendor passes all seven, the score is probably worth something. If they cannot answer question two, nothing else matters.

6. Where the score stops and the homeowner starts

A score tells you which door. It does not tell you who is behind it, and the hand-waving usually happens right here. So here is the honest boundary.

The full property record shows on every lead for free as soon as it is scanned: owner name, whether they live there or it is an absentee or rental owner, estimated market value and confidence, year built, last sale date and price, beds and baths, living square footage, lot size, stories, garage, annual property tax, and mortgage lender. Roofbird also estimates owner equity, because equity is what decides whether someone can say yes, and when the deed record cannot support an estimate it says so instead of guessing.

Only the phone and email sit behind a one-click unlock. Every number is DNC-scrubbed and labelled. Clear means it was checked against the federal Do Not Call registry and is safe for a manual sales call. DNC means do not call it. Verify means the scrub could not confirm either way, so treat it as unknown. Manual dialling only, no texts, no auto-dialler. Homeowner contact details and DNC screening covers the labelling in detail, and a lookup that finds no contact never costs a credit.

That is the payoff of a good score, by the way. The score gets you to the right block. The owner name, the equity estimate, and a clean phone number get you to the right person. A shared marketplace can hand you neither, because it does not know your territory and it sold the same lead to three other roofers before you dialled.

7. What it costs, and what it does not

One meter: credits. Looking is free, acting costs.

Asking, searching, sorting and saving lists are free and unlimited. Scanning new ground and ranking every roof in it is free at any size, and you can hold unlimited service areas. Credits are spent on reading a roof properly, which buys both photographs and the full property record together at 1 credit per 8 roofs, on the owner's name, phone and email at 1 credit per 2 roofs, and on opening a house completely at 1 credit. Roofbird pricing and what a credit buys has the exact numbers, and the full feature list covers the rest.

There is no quota and no per-lead fee. Roofs already replaced, the wrong material for your shop, or not homes at all are read, labelled as exactly that, and kept out of the way. Knowing which quarter of a street is dead is worth as much as knowing which quarter is hot. A building anyone has already assessed is never re-assessed from scratch and never re-charged.

FAQ

Q: What is AI lead scoring in roofing? A: It is software that reads a roof from imagery and public records and ranks it by how badly it needs replacing and whether anything is forcing a decision now. The useful output is a sorted list of doors, not a single number. The score is built from roof condition, roof age read off the shingles, visible active damage, roof geometry, and the condition of the neighbouring roofs.

Q: Why is a comparative rank better than a 0-100 score? A: Absolute scores cluster, so the worst roof on a block might score 68 and the best 74, which is not a ranking you can act on or repeat at a door. A comparative rank places each roof against the houses either side of it and sorts them into replace-now, worn, midlife, or already done. That is a statement you can say to a homeowner, and it is the strongest social proof in the trade.

Q: Does neighbourhood hail raise a lead's urgency score? A: It should not. Every home in the scan shares the same storm, so it cannot separate one door from the next. Hail belongs in your pitch, not in the ranking. Roofbird's NOW score uses only triggers specific to that house: visible active damage, hail damage on that roof, a recent purchase, patching already attempted, or neighbours who have re-roofed.

Q: What does confidence mean on a roof score? A: It means how much of the roof the camera could actually see. A roof behind heavy tree canopy or set far back from the road cannot be judged, and it should be labelled that way rather than mixed in with the confident ones. Where street-level imagery is newer than the aerial, the newer picture wins and both capture dates are shown.

Q: How do I test a scoring vendor before I buy? A: Pick ten houses you know, including two recently re-roofed and two original. Ask for the imagery capture date on each, check that the re-roofed houses are not at the top, look for confidence labels, and ask what the score is relative to. If the vendor cannot tell you when the photograph was taken, the score is unfalsifiable.

Next steps

1. Pick the ten houses you know best and run the seven-point check above on whatever tool you are currently using.
2. If you want to run it on Roofbird, a live scored area you can look at without signing up is the fastest way to see the ranking and the confidence labels side by side.
3. If you have a CRM full of old leads already, scoring the leads already in your CRM is often a better first move than buying new ones.
4. If you are still paying per lead, Roofbird compared with buying shared leads is the structural argument, and roofing lead generation without paying per lead goes deeper on the economics.

A score is only worth what you can verify. Verify it on your own streets before you trust it on anyone else's.