The roof tells you first

A commercial roof announces itself long before anyone inside the building picks up a phone. Ponding that never dries. Staining that follows the drainage. Patches in three different colours. All of it is visible from directly overhead, on imagery you do not need a ladder or an appointment to look at.

That is the appeal, and it is real. But most of what gets sold as “AI roof scanning” overstates what a picture can settle. Here is the honest version, including the parts that cost us money to learn.

What imagery is genuinely good at

Coarse, large-area conditions read reliably from above. Standing water shows up as dark, sharp-edged shapes that persist across captures. Widespread discolouration follows slope and drainage in a way that is hard to mistake once you have seen a few hundred roofs. Patching is obvious because the repairs rarely match. Membrane family — single-ply, modified bitumen, built-up, metal — is usually identifiable, and so is the density of rooftop equipment, which tells you something about who else has been up there.

For deciding which roofs deserve an hour of a salesperson's time, that is enough. Prioritisation is what imagery is for.

What it cannot tell you, and why that matters

Overhead imagery cannot see the substrate. Saturated insulation, fastener backout, failed flashing detail — none of it is visible from nadir. Fine defects such as seam separation and early streaking sit below the resolution of most freely available imagery, which is why a scan that confidently reports them is often reporting noise.

Two confounders cause most wrong calls. The first is vegetation and shadow: tree canopy across a roof plane reads as staining to a model that has not been taught otherwise, and the same roof photographed in March and July can produce opposite verdicts. The second is coatings. A silicone restoration coating looks bright, uniform and nearly new from directly above while the deck beneath it keeps failing — and coatings are frequently applied without a permit, so the records do not correct the picture either.

The practical rule: a clean-looking roof on an old building is a question, not an answer.

The capture date is the claim

Resolution gets the attention; the date decides whether the read is worth anything. Imagery captured before the last hail event cannot tell you anything about that event. If nobody has flown the property since the storm, the honest statement is that the roof has not been imaged since — not that it looks fine. Any condition claim that does not carry the date of the pixels it was read from should be treated as unverified.

A picture is not a lead

A worn roof with no name attached is a building, not an opportunity. The step that converts one into the other runs through the property record: the parcel gives the owner of record, the owner of record is usually an entity rather than a person, and the entity has to resolve to a human whose role actually touches the roof. Get that wrong and you pitch a tenant's office manager about a landlord's capital project.

Permit history is what turns a guess about age into a defensible sentence. A roof with a documented re-roof five years ago is not a replacement conversation — though it may be a warranty-preservation inspection if a storm has passed over it since.

What good looks like

Imagery narrows the territory. Records verify the age and the ownership. A named person with a verified phone number makes it callable. And every claim that reaches the customer carries the date and the source it came from, so the contractor can stand behind it on the roof.

That is the standard worth holding a vendor to — including us. If a tool tells you a roof needs replacing, ask what it looked at and when.

Common questions