Machine Learning Lead Engine for Commercial Roofing

The challenge

Roof-coating contractors mail whole ZIP codes from a rented list. Most of those roofs do not need coating, so the postage is wasted and the response rate stays low.

• Coating contractors buy rented mailing lists and blanket entire ZIP codes. The list has no idea which roofs are near end of life and which were just replaced, so most of the postage lands on roofs that will never convert.
• The signal that matters is on the roof itself: the material, the roof type, and how far the surface has degraded. That read exists in aerial imagery, but no contractor has a way to run it across a whole territory.
• A generic postcard gets ignored. The owner has no reason to believe the sender knows anything about their building, so even the roofs that do need coating rarely respond.

The solution

• A custom machine learning model reads every roof in the territory from aerial imagery, classifying material and roof type and scoring surface condition, with per-roof confidences shown for review rather than presented as ground truth.
• Every roof in the county is ranked on a three-axis coating-need score, so the crew works a prioritized Hot list instead of a blanket mailing and roofs that were recently replaced drop off the list.
• Each qualified roof gets a personalized mailer: an aerial crop of the owner's own building with the specific issue called out, and a coating offer that fits what the roof actually needs.

Technology stack

• Custom machine learning model (aerial roof classification)
• Aerial imagery ingestion + parcel data
• Three-axis coating-need scoring

Measurable results

• Marketing spend concentrates on the roofs that need coating, instead of blanketing rented lists where most addresses will never convert.
• The personalized, own-roof mailer gives the owner a concrete reason to respond, which lifts the odds on the same budget.
• A dealer network can hand each contractor a ranked Hot list for their own territory, so everyone knows exactly which doors to knock on.