The wrong starting point

Most vendor ROI math starts with labor. Take the loaded cost of a CSR, multiply by how many the AI replaces, and present the difference. It is a clean calculation and, for roofing, it is the small number.

Labor savings are a rounding error next to the inspections you never set. Roofing revenue is lumpy and job size is large, so a handful of additional signed contracts a month outweighs an entire salary line. Start the math where the money is.

Start from calls that never became inspections

Pull twelve months of call data and separate three buckets. Calls that rang out or hit voicemail. Calls answered but never converted to a set inspection. Calls that arrived outside business hours and got the voicemail path by default.

That third bucket is usually the largest and always the least visible, because nobody reports on calls that never reached a person. For roofers it also contains the highest-intent callers in the dataset, since a homeowner calling at 10 PM has just discovered a problem.

The calculation

Multiply steps 2 through 5. That figure is your recoverable pipeline, not your guaranteed gain, because no system converts every missed call. Take a third of it as a working assumption and the number is usually still uncomfortable.

Count the storm months separately

An annual average will understate the case badly. Roofing volume concentrates around weather, and capacity fails precisely during concentration. Run the same calculation on your two heaviest months in isolation and compare it to a quiet month.

The gap between those two results is the actual argument. Steady-state performance is not the problem you are solving. You are buying capacity for the weeks when demand exceeds anything you could reasonably staff.

What belongs on the cost side

Keep the cost side honest. Include the platform, the integration work, and the internal time to get scripts and scheduling rules right. Then include the things that usually get left off: What you currently spend on an answering service, overtime during storm weeks, and the recruiting and training cost of seasonal CSR churn.

Also check how the pricing behaves under load. Per-minute models rise with call volume, which means cost peaks in the same month revenue opportunity peaks. Ask how the invoice behaves in the month your phones triple, because in a weather-driven trade the shape of the cost matters more than the rate.

The second return nobody models

There is a return that never appears in a vendor spreadsheet because it does not look like revenue. Your CSR team currently spends its day on intake volume, which means the highest-judgment calls in the building are handled by whoever is free rather than whoever is best.

When routine intake stops competing for that attention, your strongest people work insurance escalations, upset homeowners, and the commercial callers who decide whether you win a portfolio. That shift does not show up as a line item, and operators consistently report it as the change they notice first.

The number that settles it

After 90 days, one metric answers the question: Inspections set per 100 inbound calls, measured before and after, with after-hours included in both. Everything else is proxy.

If that number has not moved, the platform is not working regardless of what a dashboard reports about answer rate or call handling time. If it has moved, multiply the delta by close rate and average job size and you have the return, calculated the same way you would evaluate a new lead source.