The 30-60-90 Day AI Plan for a Roofing Company
- Ninety days, three phases: prove one thing works (days 1-30), roll it to the team and add a second (31-60), add a third and write it all down (61-90).
- Pick ONE problem first. The companies that deploy five AI things at once end up with zero working.
- Your office manager runs this. Not you, and not the crews. Roofs get nailed the same way they did last year.
- Baseline before you start: missed calls, quote close rate, review count. No baseline, no proof, no idea what earned its keep.
- Lesson 4 of 5. Lesson 3 has the prompt library this plan uses; lesson 5 covers what to buy.
Here’s how AI adoption usually goes at a roofing company: somebody watches a demo in February, buys four subscriptions in March, and by June the only thing running is the invoice. The tools weren’t bad. The rollout was. Everything launched at once, nobody owned any of it, and there was no number to point at when someone asked if it was working.
This plan is the opposite. One problem at a time, one owner, and a baseline so the results argue for themselves. It assumes you’ve read lesson 2 (what to deploy) and lesson 3 (the prompts your office will lean on), but it works standalone.
Pick one. Just one.
The single biggest predictor of whether any of this works is whether you resist the urge to fix everything at once. Five simultaneous launches means five half-configured tools, one overwhelmed office manager, and a team that quietly goes back to doing it the old way by week three. One launch means one thing that either works or gets cut, with a number attached either way.
Roofing makes this discipline harder than most trades because the busy season compresses everything. June through August the office is triaging, not experimenting. Which is exactly why the plan is built to run on one person, part time, with the crews never touching it.
Days 1-30: prove it on live calls.
Assign a champion
One person owns this, and it's almost always the office manager, because they already touch every call, quote, and review. Not the owner: you'll get pulled onto a roof or into an adjuster meeting and the project dies in week two. The champion needs an hour a day at most.
Pick the one problem
The one costing the most money. For most roofing companies it's the phone (calls ringing out while everyone's on a roof) or the quiet quotes (estimates that never got a second touch). Pick by dollars, not by what's most fun to demo.
Baseline the numbers
Three numbers, written down before anything gets installed: how many calls went to voicemail last month, what share of quotes closed, and how many Google reviews you have next to the competitor who outranks you. Takes an afternoon. Skipping it means that in month three you'll be arguing from vibes.
Pilot on live calls
Two weeks, real customers, small scope. If the problem is the phone, turn on missed-call text-back and watch what comes back. If it's the quotes, run the follow-up prompt from lesson 3 on every estimate that goes quiet. Don't announce a transformation. Just run it and count.
Days 31-60: spread what worked.
Roll the working thing to everyone
Whatever survived the pilot goes to the whole office. Same tool, same wording, no freelancing. If the day-5 text worked in the pilot, everyone sends the day-5 text.
Add the second workflow
Now, and only now, the next problem on the list gets its turn. Usually whichever of the phone or the quotes you didn't pick first, or the review ask if you're being outranked by a company with a longer review page and worse work.
Build the prompt library
Start from the seven in lesson 3, keep the ones your office actually reuses, and add the variations that fit your jobs. A shared doc is fine. The point is that the good wording lives in one place instead of in whoever figured it out first.
Days 61-90: write it down, keep what earned its keep.
Add the third workflow
Same drill: pick it by dollars, pilot small, then spread it. Three working workflows in 90 days beats ten installed ones every time.
Write the SOPs
One page per workflow: what tool, who owns it, what the human checks before anything reaches a customer. Boring on purpose. This is the difference between a system and a thing Brenda knows how to do.
Compare against the baseline
Pull the same three numbers from day one and put them side by side. Missed calls recovered, quotes closed, reviews added. The tools that moved a number stay. The ones that didn't get cancelled without ceremony. An AI tool that hasn't moved a number in 90 days isn't early. It's furniture.
An honest note about who does the work. Nearly all of this is office-side. The crews keep roofing. If anyone tells you AI adoption means retraining the people on the ladders, they’re selling you a bigger project than you need.
Open last month’s phone log and count the calls that went to voicemail. Write the number somewhere you’ll see it. That’s your first baseline, it took ten minutes, and day one of the plan is already done.
Asked before you had to ask.
What's the right first problem to pick?
The one costing the most money, and if you don't know which that is, the baseline week tells you. Count the missed calls, count the quotes that went quiet, count the review gap. The biggest number wins. For most shops it's the phone, because a missed call in roofing isn't a missed conversation, it's a missed roof.
Do my crews need AI training?
No. This whole plan is office-side: phones, quotes, reviews, paperwork. The closest a crew gets is captioning photos the way they already do, which lesson 3's documentation prompt turns into warranty files. Nobody on a roof needs to learn anything new for this to work.
What if the 90 days lands in storm season?
Then run days 1-30 and pause there if you have to. The two busiest storm years in the county's NOAA record, through the latest available data, are 2025 (37 events) and 2024 (36), and a hail day is exactly when the office drowns in calls. A working missed-call recovery installed before the next one pays for this whole exercise by itself.
If you’d rather have someone build the first 30 days with you, walking the baseline and the pilot instead of reading about them, book 15 minutes. No deck, no discovery-call theater.