How to Vet a Roofing Property Data Provider for Accuracy

Every roofing data vendor demos beautifully. A clean map, color-coded pins, a roof age stamped next to each house, a hail swath laid over the neighborhood. It looks like certainty. Then you buy it, point a crew at the "hot" houses, and discover the 14-year-old roof was replaced four years ago, the storm date is six weeks off, and a third of the addresses route your guys to vacant lots and rentals nobody answers. The product wasn't lying. It just sold you a confidence it couldn't back up, and you didn't know which questions would have caught it before the check cleared.

I've bought, tested, and thrown out enough of these tools to know that the accuracy problem is almost never where the buyer is looking. People obsess over the map UI and the storm graphics. The money leaks out somewhere else: stale parcel records, roof-age estimates dressed up as facts, storm dates that don't survive a cross-check, and address hygiene so bad that a quarter of your mail bounces. The good news is that every one of those failure modes is testable before you pay, if you know what to ask for and how to read the answer.

What follows is the evaluation I run now, earned the expensive way. It's built around a scorecard you can reuse on any vendor, a set of trial tests that expose accuracy in an afternoon, the pricing models you'll actually be quoted (and the traps inside them), a vendor-question checklist, and a section on matching the tool to your situation so you don't buy a storm platform when you needed a list cleaner. RoofPredict shows up once, as one option among several, with its real limits stated plainly.

Why this purchase goes sideways

Roofing data is a category where the demo is the product's best moment and everything after is regression to reality. A few structural reasons that's true, and why smart operators still get burned.

The demo is curated; your territory is not. Every vendor demos on a metro where their data is fresh and dense — usually a large suburban market with good aerial coverage and clean county records. Your actual ZIPs might be rural, might be a county that digitized parcels in 2009 and stopped, might be a coastal market where re-roofs happen constantly and never hit the public record. Coverage and freshness are not uniform, and the demo is engineered to never show you the thin spots.

Roof age is the headline and the weakest link. This is the single most misunderstood number in the category. There is no national registry of roof replacement dates. When a vendor shows you "Roof age: 17 years," they derived it — from permit data, from year-built, from aerial change detection, from a model, or from some blend. Each method has a different and usually unstated error bar. A re-roof that didn't pull a permit (common) is invisible to permit-based estimates. Year-built tells you when the house was built, not when the roof was last touched. I'll go deep on this because it's where most of the accuracy conversation actually lives.

Storm data feels objective but isn't self-validating. Hail and wind reports come from real authorities — NOAA's Storm Prediction Center, the National Weather Service, radar-derived estimates. But a vendor's "this house took 1.75-inch hail on this date" is a modeled, interpolated, vendor-specific claim layered on top of that raw data. Two vendors can pull from the same NOAA feed and disagree about whether a given street got hit, because the interpolation between report points is where the modeling happens. Buyers treat the storm number as ground truth. It's an estimate with a vendor's assumptions baked in.

Address and contact data rots fast and quietly. Roughly speaking, a meaningful share of any consumer database goes stale every year as people move — the U.S. Census Bureau tracks annual mover rates in the low double digits historically. Phone numbers churn even faster. A provider can have excellent roof intelligence and still hand you a list where the deliverability is quietly mediocre, and you won't feel it until your mail-house reports the bounce rate or your callers hit dead numbers all week.

"Accuracy" is never defined in the contract. Almost no vendor will put a number on it, and the ones who do rarely define the denominator. "95% accurate" against what ground truth? Measured how? On what sample? In which markets? The word does a lot of work in a sales call and almost none in a master service agreement. Your job in vetting is to convert that vague word into something you can test and, ideally, something they'll stand behind.

The through-line: the value of roofing data is entirely downstream of its accuracy, and accuracy is non-uniform, derived, and rarely guaranteed. So you don't evaluate these tools by how good they look. You evaluate them by how well they survive a structured, adversarial check in your market.

What you're actually buying: the four data layers

Before the scorecard, get clear on what's under the hood, because "roofing property data" bundles four distinct layers that fail independently. A vendor can be excellent at one and weak at three.

1. Property / parcel layer. Who owns it, lot size, year built, square footage, owner-occupied vs. absentee, mailing address vs. site address. Sourced from county assessor and recorder records, resold and normalized by data aggregators. Freshness depends on how often the county updates and how often the aggregator re-pulls.
2. Roof condition / age layer. The roofing-specific value-add: an estimate of how old the roof is, sometimes a condition or "wear" signal from imagery. This is the most differentiated and most error-prone layer.
3. Storm / weather layer. Historical hail and wind exposure per location, derived from NOAA/NWS/SPC data and radar, interpolated to the address or parcel.
4. Contact / deliverability layer. Phone, email, and a clean deliverable mailing address. Often a separate data partnership bolted onto the property layer.

When you vet a provider, vet each layer separately. The fastest way to get fooled is to let a strong storm map convince you the contact data is good, or let an impressive roof-age demo distract you from a parcel layer that's three updates stale.

The accuracy scorecard

Here's the reusable scorecard. Score each provider 1–5 on every line for your market, not in the abstract. Anything you can't verify scores low until proven — "trust me" is a 1. I weight the roof-age and storm methodology lines heaviest because that's where the category's differentiation and its lies both live.

A quick read on scoring: a provider that lands 4s and 5s on lines 1, 2, 5, 10, and 12 is being honest with you, and honesty correlates hard with real accuracy in this category. A provider that aces the UI but stonewalls on methodology (line 1) and won't sample your streets (line 10) is selling polish. I've never regretted weighting honesty-about-limits heavily, and I've always regretted ignoring it.

The deepest line item: how to interrogate roof age

Roof age deserves its own section because it's the reason most buyers are in the market and the number most likely to be wrong. Walk through how each derivation method works and where it breaks, then I'll give you the questions and the test.

The methods, and how each one fails

Year-built as a proxy. The laziest version. The provider returns the home's construction year and implies roof age from it. Fails the instant a roof has ever been replaced — which, on a 25-to-50-year-old house, is more likely than not. A re-roofed 1968 house is not a 56-year-old roof. If a tool's "roof age" tracks year-built suspiciously closely across a sample, that's what's happening.

Permit records. Better, when they exist. A pulled re-roof permit with a date is close to ground truth. The failure modes: many re-roofs never pull a permit (homeowner-direct jobs, small contractors, jurisdictions that don't require or enforce it), permit digitization varies wildly by county, and lag between work and record can be months. Permit-based age is great where coverage is good and silently wrong where it isn't — and the tool rarely tells you which county you're in.

Aerial change detection. The vendor compares historical aerial imagery over time and infers a re-roof from a visible change in the roof surface — a color or reflectance shift between captures. Genuinely useful and increasingly common. Limits: it only sees changes within the imagery archive's window (often the last decade or so), it can confuse a cleaning, partial repair, or solar install for a re-roof, tree cover and shadow degrade it, and capture cadence varies by area. It tells you "something changed around 2019," which is a real signal, but it's a signal, not a receipt.

Modeled / blended estimates. The sophisticated providers fuse several of the above plus statistical priors (housing stock age, regional re-roof rates, material assumptions). Done well, this is the best available answer. Done opaquely, it's a black box that outputs a confident single number with no error bar. The quality is entirely in whether they'll show you the seams.

The physics reality every honest vendor will admit: roof age is a range, not a date. Anyone returning "this roof is exactly 16 years old" for every house in a county is manufacturing precision they don't have. The right answer looks like "15–20 years, moderate confidence" — a band that widens where the underlying data is thin. A vendor that gives you bands and confidence is telling you the truth about uncertainty; a vendor that gives you a clean integer for every home is hiding it.

The roof-age questions to ask

• "Walk me through exactly how you derive roof age. Which inputs, in what priority?"
• "Is the output a single year or a range? What's the typical error band, and where does it widen?"
• "In my counties, which method is doing the work — permits, imagery, or model? How's your permit and imagery coverage here specifically?"
• "How do you handle a re-roof that didn't pull a permit?"
• "What's your imagery capture cadence in my area, and how far back does the archive go?"

If the rep can answer these without retreating to "it's proprietary," you're talking to a real data company. If every answer is a black box, score line 1 a 1 and adjust your trust accordingly.

The roof-age ground-truth test

This is the single most valuable thing you can do, and almost nobody does it. You already have ground truth — your own job history.

1. Pull 25–40 addresses from your own records where you know the truth: roofs you replaced (you know the exact install date), roofs you inspected and confirmed were newer, and a few you know are original and aging.
2. Hand those addresses to the vendor during the trial and have them return roof age for each — without telling them which is which.
3. Score it. For the roofs you replaced, does their estimate land in the right neighborhood, or does it still say "18 years" on a roof you put on three years ago (which would mean they're missing re-roofs — the cardinal sin)? For the original roofs, are they appropriately flagged old?
4. Note the direction of the errors. Systematically too old means they miss re-roofs and you'll waste trips on good roofs. Systematically too young means they'll skip houses that are actually due. Both are bad; knowing which one tells you how the tool will fail you in production.

A vendor confident in their data will welcome this test. One that gets squirrely about a blind accuracy check on addresses you control is telling you something. I have walked away from deals purely on the reaction to this request.

Interrogating the storm layer

If you do storm/restoration work, the weather layer is doing real work in your targeting, and it's the second-most-likely place to get fooled because the underlying authority data is legit but the per-house claim is modeled.

Provenance first

Ask where the raw data comes from. A credible answer names NOAA, the National Weather Service, the Storm Prediction Center's storm reports, and radar-derived hail/wind estimates. Those are the real authorities, and their report databases are public. If a vendor can't or won't name the source — "it's our proprietary weather data" with nothing underneath — treat the storm layer as unverified.

Then ask the question that separates serious tools from hail-map wrappers: how do you get from sparse storm reports to a per-address exposure number? Storm reports are point observations — a trained spotter or a radar pixel says hail of size X near location Y at time Z. The space between those points is interpolated, and that interpolation is the entire ballgame. A thin tool just drapes the official swath polygon over a map and calls every house inside it "hit." A better tool models the gradient and gives you a per-location intensity with some honesty about confidence. The distinction between "where the storm passed" and "which roofs it actually wore out" is exactly where vendors differentiate — and where some overclaim.

The storm cross-check test

You can verify storm claims against free public sources in minutes, which is more than you can say for roof age.

1. Take three or four specific dated storm events the vendor flags in your market.
2. Cross-check each against NOAA/SPC public storm reports and NWS records for that date and county. Do the dates line up? Is the reported hail size in the same range the vendor claims?
3. Now check resolution. Pick two streets a mile apart that you know took different damage in a real storm. Does the tool distinguish them, or does it paint the whole ZIP one color? A tool that can't tell apart streets you know differed is operating at ZIP resolution and selling you address resolution.
4. Check the date precision. "Recent storm activity" is useless for restoration work; you need the specific event so you can speak to it and document it. Vague time language is a tell that the storm layer is thin.

One honest caveat to keep your own expectations calibrated: a storm forecast or even a confirmed storm exposure is odds, not proof of damage. A roof in a 1.5-inch hail zone is more likely to have damage, not certain to. Any vendor implying their storm layer tells you which roofs are damaged (rather than which were exposed) is overselling. Exposure plus age plus an actual inspection is how you get to damage. The data narrows where to look; the ladder confirms it.

Address and contact deliverability: the unglamorous layer that eats budgets

The storm map gets the demo time. The deliverability layer gets your money when it's bad. Direct mail and call campaigns live or die on whether the address is real and the contact reaches a human, and this layer rots constantly.

What good hygiene looks like

For mailing addresses, the baseline is USPS-grade processing: CASS-certified address standardization (so the address is in the form USPS can actually deliver) and NCOALink move-update (so you're not mailing people who moved). These are real, named USPS programs; a serious data provider runs them or partners with someone who does. If a vendor can't tell you whether their addresses are CASS-processed and move-updated, assume they aren't, and assume your bounce rate will show it.

For contacts, ask for the match rate (what share of properties they can attach a phone or email to) and the recency of those contacts. A 40% phone match of recently-verified numbers is more useful than a 90% match of numbers last seen in 2017. Phone data decays fast; the age of the data matters as much as the coverage.

The deliverability test

1. Take a known sample — a neighborhood you've worked, where you know which houses are owner-occupied, which are rentals, which are vacant.
2. Run the vendor's list against it. How many absentee/rental flags are right? How many addresses resolve to real, deliverable mailing addresses?
3. If you can, run a small physical test mailer or a calling session on a few hundred records and measure the bounce/dead-number rate yourself. The mail-house's bounce report is ground truth you can't argue with.
4. Spot-check owner-occupied flags against what you know on the ground. A wrong absentee flag means you're either skipping a homeowner who'd buy or mailing a landlord who won't.

A tight deliverability layer quietly saves you more money than a flashy storm map makes you, because waste compounds across every campaign. This is the line item buyers underweight and regret.

How to run the trial so it actually proves something

Most buyers "trial" a tool by clicking around the UI for twenty minutes and reacting to whether it feels good. That tests the interface, not the data. Run the trial like an experiment.

Pick your hardest market, not their best. Trial the tool where you need it to work — your real territory, including the rural or older counties. If it only works in dense fresh-data suburbs, you need to know that now.

Bring your own ground truth. The roof-age test, the storm cross-check, and the deliverability test above all depend on you supplying addresses where you already know the answer. Your job history is the most valuable test asset you own. Use it.

Blind the test where you can. Don't tell the vendor which addresses are your re-roofs. You want to see what the tool says cold, not what the rep can explain after the fact.

Measure direction, not only magnitude, of error. As noted, "too old" and "too young" fail you in opposite ways. A tool that's consistently slightly conservative is workable; one that randomly swings both directions is not.

Take it all the way to the field on a small batch. The only complete test is doors knocked or mail sent on a small, controlled batch, tracking contact rate, appointment rate, and whether the roofs were actually the age the tool claimed. Numbers below are illustrative placeholders to show the shape of the scorecard, not benchmarks to expect — your market will differ:

Fill that table with your own numbers and the decision usually makes itself. The point isn't the specific figures — it's forcing every vendor through the same concrete tests on the same known ground so you're comparing data, not demos.

Coverage and freshness: the question the demo never raises

Two properties can look identical on a vendor's map and be worlds apart in data quality, and the difference is coverage and freshness — how complete the underlying source is in your area, and how recently it was pulled. This is the quietest accuracy killer because the UI renders a confident pin whether the data behind it is from last month or from 2018.

Coverage is geographic, not national

Vendors love the phrase "national coverage," and it's nearly meaningless for accuracy. Coverage is the share of properties in a specific county for which the vendor actually has good parcel, imagery, and permit data — and that varies street by street, not merely state by state. Aerial imagery archives are denser and more frequently refreshed over large metros than over rural fringes. Permit digitization is excellent in some counties and barely exists in others. A vendor with true national reach still has thin and thick zones, and the demo always shows you a thick one.

The way to test it: ask the rep, point-blank, "Where is your data thinnest in my market, and how would I know when I'm standing in a thin area?" A vendor who can answer honestly — "these two rural counties are weaker on imagery cadence, and the roof-age confidence drops there, which you'll see in the wider band" — is one that built coverage awareness into the product. A vendor who insists coverage is uniformly excellent everywhere is either not measuring it or not telling you. Then verify by running your known-address tests in your worst county, not your best one. If the tool only performs in dense suburbs and falls apart in the rural ZIPs you actually work, you need that surfaced before you sign, not after.

Freshness is a date, and you should be able to see it

Every layer has a vintage — a date it was last sourced. Parcel data is only as current as the last county update and the last time the aggregator re-pulled it. Imagery is only as current as the last flight or satellite pass over that area. Permit data lags the actual work by months. A serious provider can tell you, per layer, the "as of" date and the re-pull cadence. A provider that has no answer for "how old is this data?" is one where you're flying blind on the most important quality dimension there is.

The practical risk freshness creates is staleness-driven error that grows over time. A roof-age estimate that was accurate at pull is a year more wrong a year later if the data never refreshes — a roof that was "14 years" is now 15, and any re-roofs done since the pull are completely invisible. If you're paying a subscription, confirm the data refreshes on a schedule and that you're not looking at a frozen snapshot from your onboarding date. If you're paying per-record, understand that records you bought months ago are decaying, and budget for refreshes rather than assuming a one-time pull stays good.

A blunt rule of thumb I use: treat any number without a visible "as of" date as suspect, and treat any vendor who can't produce that date as one whose freshness you have to assume is bad. The honest ones surface vintage in the product; the rest hide it because it's not flattering.

Exportability and integration: don't trap your data

The accuracy of the data matters only if you can get it into the tools your team actually uses. A provider can have the best roof-age model in the category and still be the wrong choice if the data is locked in their UI and your reps have to retype addresses into the CRM by hand. Evaluate the plumbing, because it's where good data goes to die.

Clean export. Can you pull the data out as a usable file (CSV or similar) without jumping through hoops? If the only way to use the data is inside the vendor's own interface, you're not buying data — you're renting a screen, and you'll lose everything the day you leave.

CRM sync, one-way or two-way. Most shops run on a CRM — HubSpot, ServiceTitan, JobNimbus, AccuLynx, Jobber, Roofr, SalesRabbit, CompanyCam, and others. Ask which CRMs the provider syncs with, and crucially whether it's one-way (data pushed in once) or two-way (status and outcomes flow back, so your targeting learns from results). Two-way sync is what lets you eventually measure whether the "hot" houses actually converted — which is the only real long-run accuracy check. A vendor with no integration story means a permanent manual re-entry tax on every campaign.

Field-app and canvassing handoff. If you knock doors, the data has to land in whatever your reps carry to the door. A ranked list that lives only on an office desktop doesn't help the rep standing on the porch. Confirm the path from the data to the field before you assume it exists.

The test here is simple: during the trial, actually export a batch and actually push it into your CRM. Don't take "yes we integrate" on faith — wire it up on the trial data and see whether it's clean or a mess of mismatched fields. Integration demos are notoriously optimistic; the trial is where you find out if it's real.

Pricing models you'll be quoted, and the traps in each

The category prices a few different ways, and each has a characteristic trap. Knowing the model tells you where to push.

Per-record / per-report. You pay for each property record or report pulled. Trap: the price-per-record is meaningless if accuracy is low — cheap wrong data is the most expensive data you can buy, because you pay again in wasted trips and mail. Also watch for whether refreshes cost again; a record you pulled six months ago is now stale and you may be re-buying it.

Monthly subscription / seat-based. Flat monthly access, sometimes per user. Trap: usage caps and overage fees buried in the agreement, and "unlimited" plans that throttle or rate-limit in practice. Confirm the cap and the overage rate in writing.

Territory / market-area licensing. You license a geography. Trap: exclusivity language. Some vendors imply exclusivity in your area; get specifics on whether competitors can license the same territory, and don't pay an exclusivity premium for a promise that isn't in the contract.

Done-for-you bundles (data + mail or data + leads). The data is bundled with a mailing service or lead delivery. Trap: you lose the ability to inspect the data layer independently, and the margin is hidden inside the service. Also be clear-eyed about whether you're buying data (which sharpens your own outreach) or leads (a fundamentally different purchase — you're renting a homeowner who may be sold to several competitors). Those are different products with different economics; don't let a bundle blur them.

Across all models, the questions that protect you: What's the total cost at my real volume, including refreshes and overages? What am I locked into — month-to-month or annual? Can I export my data if I leave, or is it hostage in the UI? Is there a real trial, or just a demo? A vendor confident in retention offers easy exits; lock-in clauses are a tell that the product expects you to want out.

The vendor-question checklist

Print this and run it on every provider. The goal is to convert sales-call vagueness into testable, on-the-record claims.

On methodology and accuracy

• How exactly is roof age derived, and is the output a single year or a range with confidence?
• What's the typical error band, and in which conditions does it widen?
• When you claim an accuracy figure, accuracy against what ground truth, measured how, on what sample, in which markets?
• How do you handle re-roofs with no permit?

On data sources and freshness

• What's the source and "as of" date for each layer — parcel, roof age, storm, contact?
• How often do you re-pull each layer, and how does that vary by county?
• For storm: do you source from NOAA/NWS/SPC, and how do you interpolate to the address?

On coverage

• Show me confirmed coverage and freshness in my specific counties, not a national map.
• Where is your data thinnest, and how would I know when I'm in a thin area?

On deliverability

• Are mailing addresses CASS-certified and NCOALink move-updated?
• What's the phone/email match rate and how recently were those contacts verified?
• How reliable is the owner-occupied vs. absentee flag?

On commercials and control

• Total cost at my real monthly volume, including refreshes and overages?
• Real trial on my territory with my addresses — yes or no?
• Can I export my data and integrate with my CRM? Which CRMs sync, one-way or two-way?
• What's the exit — month-to-month, or am I locked into a year?

The two questions that reveal character

• What can your tool not do well, and who is it wrong for?
• Can I run a blind accuracy test on 30 addresses I already know the truth about?

The last two are the tells. Honest data companies answer the limits question without flinching and welcome the blind test. The reaction to those two questions has predicted vendor quality for me better than any feature list.

Match the tool to your situation

There is no single best provider — there's the right one for how you actually work. A few common situations and what to weight.

You mail and want to stop wasting postage on new roofs. Your make-or-break layers are roof age (so you skip new roofs) and deliverability (so the mail lands). Weight scorecard lines 1, 2, 3, 7, and 9 heaviest. A gorgeous storm map is irrelevant if you're not doing restoration; don't pay for it.

You knock doors and want your crew on the right streets. You need roof age plus a clean ranked list and solid owner-occupied flags, and you need it to export to whatever your reps carry. Weight lines 2, 9, 10, and 11. Street-level accuracy matters more than national breadth — you only knock where you knock.

You do storm restoration. The storm layer earns its keep here, but only if it's address-resolution and provenance-clean, paired with roof age so you're not chasing new roofs that took hail and don't need replacement. Weight lines 5, 6, 1, and 2. Run the storm cross-check test hard, and remember exposure is odds, not damage.

You sit on a big CRM book and old estimates. Your highest-ROI move may not be new property data at all — it's re-scoring the homeowners already in your database (old estimates, past customers) by current roof age and recent storm exposure. For this, prioritize a provider that ingests your list and enriches it, and weight integration (line 11) and roof-age freshness. Buying a fresh territory when you haven't mined your own book is spending money to ignore money you already have.

You're a smaller shop and price-sensitive. Avoid annual lock-in and rich bundles; favor per-record or month-to-month so you can test cheaply and walk if accuracy disappoints. The blind ground-truth test matters even more for you because you can't afford to learn the data is wrong in production.

A worked example: vetting two providers in an afternoon

To make the process concrete, here's the shape of a real evaluation — names omitted, numbers illustrative — so you can see how the pieces combine into a decision.

Say you run a residential shop in an older inner-ring suburb with a mix of original 1960s–80s roofs and a steady churn of re-roofs that mostly didn't pull permits. Two vendors are in the running. Both demo well.

1. You pull 30 known addresses from your job history — 10 you re-roofed in the last 5 years, 10 you confirmed newer on inspection, 10 you know are original and aging — and hand both vendors the bare addresses.
2. Vendor A returns ranges with confidence and correctly flags 9 of your 10 re-roofs as newer; Vendor B returns clean single-year numbers and still calls 6 of your 10 re-roofs "18+ years." That's the whole ballgame on roof age: B is permit-leaning in a no-permit market and missing re-roofs systematically. In production, B would march your crew onto good roofs all day.
3. You cross-check storm claims against SPC public reports for three dated events. A matches all three and distinguishes two streets you know took different hail; B matches two and paints the ZIP one color.
4. You run a small test batch through your mail-house. A's addresses come back cleanly deliverable with accurate absentee flags; B's bounce rate is meaningfully higher and flags two known rentals as owner-occupied.
5. You ask both the limits question. A's rep volunteers that their imagery archive only goes back about a decade so very old re-roofs are fuzzier, and that a couple of your rural fringe ZIPs are thinner. B's rep insists the data is "99% accurate" and dodges the blind-test framing.

The scorecard isn't close, and notice it had almost nothing to do with which UI looked nicer. It was decided by methodology honesty, a blind ground-truth test, a free public cross-check, and a small real-world batch. That's the entire method. You can run it on any vendor in the category.

Where RoofPredict fits — one option, with limits

Since this is a category buyer's guide, here's an honest placement of one tool I know. RoofPredict ranks roofs by age band and storm exposure, house by house, so a contractor can point mail, door-knocking, or an old CRM list at the homes most likely to be due, and skip the new ones. It models hail and wind exposure per roof rather than draping a ZIP-level swath over the map, and it's built to enrich your own list as well as sell you a new territory — which fits the "I have a CRM book" and "I mail and waste postage" situations above.

The limits, stated plainly because that's the whole point of vetting: roof age comes back as a range, not an exact date (anyone promising an exact year is overpromising); the storm exposure is odds, not proof of damage — it tells you which roofs were likely worn, not which are definitively damaged, and an actual inspection still confirms the job; and the scoring is roof-age and storm-exposure heuristics, useful targeting, not a magic condition reading of every shingle. It is not a measurement tool — if you need precise roof measurements for estimating, that's a different category (EagleView, HOVER, Roofr and the like measure the roof; they don't tell you which roof is due). And it's a newer entrant, so run the same blind ground-truth test on it that you'd run on anyone. Put it on the scorecard next to the others and let the tests decide.

The broader point stands regardless of which tool wins your evaluation: roofing property data is only worth what its accuracy is worth in your market, accuracy is derived and non-uniform, and every failure mode above is testable before you pay. Run the scorecard, bring your own ground truth, cross-check the storm claims against the public record, and weight honesty about limits as heavily as anything on the feature list. The vendors who survive that are the ones worth your money — and you'll know exactly why.

The 30-minute pre-purchase checklist

If you do nothing else, do this before you sign anything:

1. Pull 30 addresses from your own job history where you know the truth.
2. Make the vendor return roof age on them, blind, as a range.
3. Count the missed re-roofs — that one number tells you most of what you need.
4. Cross-check three dated storm claims against NOAA/SPC public reports.
5. Confirm mailing addresses are CASS-certified and move-updated, and run a small batch through your mail-house.
6. Confirm coverage and freshness in your counties, not a national map.
7. Ask the two character questions — "what can't it do?" and "can I run a blind test?" — and watch the reaction.
8. Get total cost at your real volume and the exit terms in writing.

Do that, and you'll have replaced a confident demo with evidence — which is the only thing that's ever protected me from buying expensive certainty I couldn't back up.

FAQ

How accurate is roof age data, really?

It varies enormously by provider, method, and your specific market — and it should always be a range, not an exact date. There is no national registry of roof replacement dates, so every "roof age" is derived from permits, year-built, aerial change detection, or a blended model, each with a different error band. The honest answer is that good roof-age data is directionally reliable for targeting (skip new roofs, find old ones) but is not a precise per-roof condition reading. Test it on your own known roofs before trusting it.

What's the single most important test before buying?

The blind ground-truth test on roof age. Pull 25–40 addresses from your own job history where you know the real roof age — especially roofs you replaced — and have the vendor return age for them without telling them which is which. Count how many re-roofs they miss (call old when you know they're new). That one number predicts how the tool will waste your crew's time in production better than any feature in the demo.

How do I verify a provider's storm and hail data?

Cross-check it against the public record. NOAA, the National Weather Service, and the Storm Prediction Center publish storm reports you can look up by date and county. Take three or four specific dated events the vendor flags and confirm the dates and hail sizes line up. Then test resolution: pick two streets you know took different damage in a real storm and see whether the tool distinguishes them or paints the whole ZIP one color. A tool that can't tell apart streets you know differed is operating at ZIP resolution.

Why does roof age have to be a range instead of an exact year?

Because no one has the receipt. Roof replacements aren't recorded in any complete national database, and many re-roofs never pull a permit, so providers infer age from imperfect signals. An honest tool returns a band like "15–20 years, moderate confidence" and widens that band where the underlying data is thin. A provider that stamps a clean single year on every house is manufacturing precision it doesn't have — which is a reason to trust it less, not more.

What address-hygiene standards should a provider meet?

At minimum, mailing addresses should be CASS-certified (standardized to a USPS-deliverable format) and NCOALink move-updated (so you're not mailing people who moved). These are named USPS programs a serious data provider runs or partners for. If a vendor can't tell you whether their addresses are CASS-processed and move-updated, assume they aren't — and assume your bounce rate will prove it. Run a small test batch through your mail-house and read the bounce report yourself.

Is buying roofing property data the same as buying leads?

No, and conflating them costs money. Property data sharpens outreach you already do — it tells you which of your own doors, mailers, or CRM contacts to prioritize, and you own that work. Leads are a different product: you're typically renting a homeowner's interest, often sold to several competitors at once. Some vendors bundle the two, which hides the data quality inside a service. Decide which one you actually need before evaluating, and don't let a bundle blur the line.

How should I weight the scorecard if I only do storm restoration?

Weight storm provenance and resolution heaviest, but never drop roof age. The storm layer should cite NOAA/NWS/SPC sources, give you specific dated events, and resolve to the street or address rather than the ZIP. Pair it with roof age so you're not chasing new roofs that took hail and don't need replacement. And keep expectations calibrated: storm exposure is odds, not proof of damage — it tells you which roofs were likely worn and worth inspecting, not which are definitively damaged.

What pricing traps should I watch for?

Per-record pricing looks cheap but cheap wrong data is the most expensive data you can buy, and you may re-pay to refresh stale records. Subscriptions can hide usage caps and overage fees — confirm both in writing. Territory licensing can imply exclusivity that isn't actually in the contract. And done-for-you bundles hide the data margin inside a service so you can't inspect the data layer alone. Always ask for total cost at your real volume including refreshes, and get the exit terms in writing.

How do I test a provider in my market instead of their demo market?

Insist the trial runs on your actual territory, including your rural or older counties, not the dense fresh-data metro they demo on. Bring your own ground-truth addresses for the roof-age and deliverability tests, cross-check storm claims against public NOAA/SPC data, and take a small batch all the way to mail or doors to measure real contact and bounce rates. If a vendor will only show you their cherry-picked metro and won't sample your streets, score that against them — it usually means their coverage is uneven.

Which two questions actually reveal whether a vendor is trustworthy?

First: "What can your tool not do well, and who is it wrong for?" Honest data companies answer without flinching; the ones who claim they do everything perfectly are the ones to worry about. Second: "Can I run a blind accuracy test on 30 addresses I already know the truth about?" A vendor confident in their data welcomes it; one who gets squirrely is telling you something. The reaction to those two questions has predicted vendor quality more reliably than any feature list.