How a regional roofing company reclaimed their nights with an AI receptionist
Mountain Ridge Roofing runs two crews across western North Carolina. They were answering every call themselves — during dinner, during school pickup, during the one weekend they tried to take off as a family. Most of those calls were the same few questions: Do you service Boone? What's a typical turnaround? Can someone come look at a storm leak tomorrow? The owners weren't avoiding the calls. They were drowning in them.
This is the case study of how LoopLogic replaced that always-on burden with an AI phone receptionist, and what changed in the eight weeks after it went live.
The starting point
The numbers were uncomfortable but unsurprising. Three of them did most of the talking.
Of every call the owners picked up, 62% fit a known pattern — service-area check, insurance-active storm leak, or "can I get on the schedule next week."
>
71% of inbound calls landed outside business hours — evenings, weekends, and the morning before anyone was at a desk.
>
On a 14-day trace, average time-to-first-response was 41 minutes, because every reply went through someone who was already mid-task.
The owners had three jobs and a two-person office. Each inbound call interrupted one of them. Even the ones that didn't book work still cost the day.
What we built
LoopLogic didn't ship a generic chatbot. The system was tuned to the way Mountain Ridge actually quotes and books work.
Voice + intake
Calls route to a Vapi-powered AI receptionist that answers in under two rings. The system prompt is the company's own — pricing language, the local terms for roof types, the friendly phrasing the owners use on the phone. It asks the questions the lead form asks, in the order the owners themselves would. It's trained to recognize hail/storm reports, insurance-active leaks, and "just shopping around" — then transcribe the whole call and score it for urgency.
Follow-through
The receptionist doesn't just take the message. Each call lands in the CRM as a `Lead` row with the transcript, the extracted intent, the urgency score, and a suggested next-step. The owners get one email per qualified call, with a link to the dashboard to confirm a slot. The same CRM row drives the SMS follow-up ("Thanks for calling — Mountain Ridge will be in touch tomorrow morning") and the check-in a week later ("How did the inspection go?").
Rules we negotiated up front
- The AI never quotes a price over the phone. It captures the job, books the inspection, and the owner confirms the estimate on-site.
- Anything that smells like an insurance-claim dispute gets a hand-off flag. LoopLogic's rule layer bumps those rows to the top of the morning queue.
- The owners can pull the live transcript from the dashboard at any time and call the customer back. Nothing is hidden.
The first eight weeks
Eight weeks after launch, the dashboard told a clear story.
| Metric | Before | After |
| Calls answered within 30s | 38% | 96% |
| Time to first response (median) | 41 min | 3 min |
| After-hours bookings / week | 1–2 | 7–8 |
| Quotes sent within 48h | 51% | 88% |
The AI booked 31% of the company's weekly appointments — and almost all of those were after-hours calls that used to go to voicemail.
What the owners changed about their own day
The most interesting number isn't the booking rate. It's the one the owners reported on a feedback call in week six:
"We started taking our kids to school again."
That's not in the dashboard. It's the part that makes the system worth building.
What didn't work
Two things worth saying out loud.
The system prompt was too long at first
The first version tried to teach the AI the full pricing sheet. Calls dragged past four minutes and customers hung up. We cut the prompt in half, moved pricing to the dashboard handoff, and average handle time dropped from 4m 12s to 1m 48s. Less script, more triage.
Not every after-hours call should be booked
Early on, the system scheduled an inspection for a homeowner who just wanted a price on a full re-roof. The owners drove 40 minutes to a meeting that wasn't ready to quote. The fix was a single scoring rule: any call without a named problem or urgency signal goes into a "follow up Monday" queue instead of onto the calendar. That cut wasted trips to two in eight weeks.
What we'd build next
The system has held up. The next phase is review generation — a short, opt-in SMS sequence that goes out two weeks after a completed job, tied to the same CRM row. Then a monthly digest email so the owners can read what's happening without opening the dashboard.