AI Roofing Quote Waste Factor: Fixing Ice and Water Shield Math in a Live Test

An AI quote can look finished and still order the wrong amount of material. When we ran a roofing quoting assistant live against a real measurement report, its math missed on ice and water membrane, and the owner's field knowledge exposed the gap on the spot.

The company is a roofing, siding and custom-home business in the Midwest. Here is what we tested, what was wrong, how we fixed it, and what is still open. If you are building a quoting tool for your own crews, roofing contractors will recognize every line item.

The situation

We gave the assistant a real aerial measurement report and had it show how it got each quantity: roof squares, starter, drip edge, ridge cap and membrane. Then we compared its numbers with a finished estimate that the company had already produced by hand.

The last step mattered most. We asked the assistant to list every discrepancy between its version and the finished one, and to check its own work. The owner then explained why each gap existed.

What was broken

Four logic gaps came out of that one walkthrough.

• Ice and water membrane: the assistant counted one row along the eaves. The owner explained that with an overhang, one row does not reach the required distance into the living space, so eaves need two rows. Valleys stay at one row.
• Waste factor: membrane was running at the 12% baseline from the earlier test. Crews tend to throw away small leftover pieces and wrap over hips, so the owner wanted a higher number.
• Starter and drip edge: these have to come from eaves and rakes together, not eaves alone.
• Labor quantity: it did not match the shingle quantity. The owner's rule is that install squares equal shingle squares, plus starter bundles divided by three, plus ridge cap bundles divided by three.

What we changed

We fixed the two-row requirement for eaves and raised the waste factor on membrane to 20%. We also added the starter, drip edge and labor rules to the instructions.

Then we cleaned up all the existing PDF files the assistant works from, so the corrected rules replace the older logic instead of sitting beside it. Finally, we split the system into two dedicated assistants, one for roofing quotes and one for siding quotes.

Why siding became its own assistant

In the owner's experience, aerial measurement reports are unreliable for siding quantities. That makes siding a different problem from roofing. The siding lead will teach the takeoff logic, so the siding assistant learns from the person who does those takeoffs.

One question was parked on purpose. The report's estimate of attic space was not close enough to rely on, so the ventilation calculation needs its own method and was not settled on this call.

What changed

No result was reported, because the fixes had only just been made. What changed was the assistant's logic and structure. It now handled membrane rows, waste, starter, drip edge and labor the way the company does, and roofing and siding no longer shared one set of instructions.

A follow-up checkpoint was set for the next day to verify the fixes, ahead of the team using the assistant on live quotes.

What to take from this

This is advice for your own testing. Use a finished job as the answer key, make the assistant show its work line by line, and have your most experienced estimator explain every mismatch. The rules that surface this way are often the ones new salespeople get wrong.

Frequently asked questions

Why does an AI roofing quote need a waste factor?

Material is ordered in whole rolls and bundles, and crews lose some to cuts and offcuts. A waste factor adds that loss to the order. If it is too low, the crew runs short mid-job. In this case the factor on membrane was raised to 20%.

Why does ice and water shield sometimes need two rows?

The owner explained that the membrane has to reach a set distance into the living space, and an overhang pushes the edge of the wall further from the eave. One row may not get there, so eaves use two rows. Valleys stayed at one row in this case.

Should roofing and siding use the same AI assistant?

Not here. Siding quantities need different takeoff logic, and measurement reports were seen as weak for them. A dedicated siding assistant, taught by the person who does the takeoffs, keeps each assistant focused and easier to correct.

How do you find logic gaps in an AI quoting tool?

Run it on a real report, make it show its working, and compare the result with a finished quote. Ask it to list every discrepancy, then have an experienced estimator explain each one. Each explanation becomes a rule you can add and retest.

If your quoting tool has never been checked against a finished job, book a 30-minute AI follow-up and reporting audit. We walk through where your material math can drift and how to test it.