What "AI roof measurement" actually means: the five jobs AI does, and the one it still cannot

"AI roof measurement" is a label for five separate jobs: drafting the roof outline, setting the scale on a photo, cleaning up the photo, naming the edges, and suggesting a waste percentage. Some of those are machine-learning models, some are plain geometry or rules sold under the same label, and none of them can read the pitch of a roof from a straight-down image.

That last point matters most, because pitch drives every square on the order. Here is each job, what it saves you, and what you still check.

The quick test before you trust any "AI" feature#

Put four checks to any tool, ours included:

• The output. A polygon, a scale, a picture, a set of labels, or a number.
• Shown work. You should see what it changed before it applies anything.
• One-step undo. Without it, you spend longer fixing the result than doing it by hand.
• The pitch. If a feature claims to know pitch from overhead imagery alone, ask where that number came from.

Job 1: drafting the roof outline#

This is what most people picture: point a model at the image and it draws the facets. When the roof is a clean gable on a clear capture, a drafted outline is a fast start. On a cut-up roof with trees over one side, a dormer and a porch tie-in, the draft is a sketch you redraw.

RoofStruct does not draft outlines for you today; you trace the roof on satellite imagery or on your own drone or phone photo, in minutes for a typical house. Whatever tool drafts it, the check is the same: does every facet close, does the eave line follow the drip edge rather than the wall, and did it split the roof where the planes actually change.

Job 2: setting the scale on your own photo#

Satellite imagery arrives with a scale. A drone or phone photo does not, and a wrong scale is wrong on every number in the report. In RoofStruct, "Find the scale for me" works on custom uploads only: it spots a car, a garage door or a scale bar in the photo and proposes a scale. It costs 1 AI credit per run, and you confirm the number before it applies.

Treat the result as an estimate. About ±10 % is the inherent limit of reading scale from an everyday object, because cars and garage doors vary. If you can, check it against one dimension you taped on site. The full drone workflow is in how to measure a roof from a drone photo.

Job 3: cleaning up the photo#

A soft, hazy drone shot makes the hips and valleys hard to place. AI enhancement sharpens your own uploaded photo (drone, phone or blueprint) in place, with the same geometry, so a traced line lands where it would have on the original. It never runs on satellite imagery. There are three levels: Sharper at 1 credit, Studio at 2, Pro at 3.

What enhancement cannot do is invent detail that the camera never caught. If a valley is under a tree in the original, it is under a tree after.

Job 4: naming the lines#

Once the facets are traced, every line needs a type: ridge, hip, valley, eave, rake, and the rest. This is where a lot of "AI" is really geometry. RoofStruct's Auto-Label Edges button is a good example, and we say so plainly: it is not a trained model. Two geometry solvers each work out what an untyped line should be (ridge, hip, valley, eave or rake) and the button types only the lines both agree on. Disputed lines stay yellow for you to decide, lines you typed by hand are never changed, and the whole pass is one Undo step.

It passes the quick test: it shows its work and undoes in one step. Before you order, still read the "Edges labeled X of Y" count and look at every yellow line.

Job 5: suggesting a waste percentage#

A suggested waste figure is usually rules applied to what you drew, not a model. Smart Waste, on Pro, recommends a percentage from the roof's facets, pitch and valley count, and you can adjust it per job. A suggestion is only as good as the drawing under it: miss a valley and the waste comes out low. Our waste factor guide by roof type covers the cuts that get forgotten.

The one job it still cannot do: pitch from above#

A straight-down image flattens every plane. A 4/12 and a 10/12 roof with the same footprint look identical from overhead, yet the 10/12 has about 24 % more roof area (pitch factors 1.054 against 1.302). No model can recover a measurement the image does not contain.

So pitch comes from somewhere else: a gable-end photo, a pitch gauge on site, the plans, or a side view. The free pitch finder puts 12 pitch lines over a Street View photo so you can match the slope by eye; it is a visual aid and you confirm the pitch. The background is in finding roof pitch from satellite imagery and Street View.

RoofStruct also refuses to guess: roof area only appears once every face has a pitch. Until then the totals show the footprint and a "needs a pitch" note.

Where you trip#

• Trusting a drafted outline on a complex roof. Redraw the valleys and tie-ins yourself; that is where drafts go wrong and where the order gets expensive.
• Skipping the scale check on an upload. One wrong scale moves every line and every square. Tape one wall or one garage door.
• Reading "AI" as "measured". An assistant that writes a confident roof area without a trace is guessing; we walked through one in can ChatGPT measure a roof.
• Ignoring the yellow lines. Untyped or disputed edges mean missing drip edge, starter or valley metal on the order.
• Letting a default pitch stand. If every face inherited the same pitch, check the porch and the dormers.

FAQ#

Is AI roof measurement accurate?#

It depends on which job. Scale from a car or garage door is an estimate within about ±10 %. Enhancement does not change geometry. Labels and waste are only as accurate as your trace. Pitch has to come from a side view or the site.

Do I need AI to measure a roof?#

No. Tracing on satellite imagery and labelling the edges by hand works on any roof. The AI features save time on uploads and photos; they do not replace the trace.

Can AI tell the pitch from a satellite image?#

Not from a straight-down image alone, because the plane's slope is not visible from directly above. Use a gable-end photo, the pitch finder or a gauge on site.

What if I don't want to draw at all?#

Order a done-for-you report. A technician measures it from high-resolution satellite imagery and a second technician checks it; reports are usually delivered within 4 hours (turnaround can vary with volume and complexity).