April 14, 2026 · Josh Shames · 7 min read
The full breakdown. The leak we found. What we deployed. What it cost. What it earned back month by month. Client-anonymized until they decide to go on the record.
Second-generation family-run restaurant group. Three locations across one metro. Mix of sit-down dinner service, weekday lunch traffic, and a growing but undermanaged catering business. Annual revenue around $4.2M. Owner does his own books, his own staffing, and — this was the problem — his own phones.
Catering inquiries were supposed to be 15% of revenue, based on what his dad used to do. They were 6%. We pulled the call log data from his phone system in the free AI Calibration Call. The pattern was obvious within 30 seconds.
Between 5pm and 9pm — the exact window when corporate catering coordinators and event planners were placing inquiry calls — the main-location phone rang an average of 14 times per weekday. About 9 of those went to voicemail. Of the 9, about 2 got called back the next day. The rest sat. Three to four inquiries per week evaporated because nobody picked up.
At an average catering ticket of roughly $900 and a 40% close rate on inquiries that got a real conversation, he was losing somewhere around $4,300/month in booked business. $51,600/year. On one leak.
Phase 1 was not a redesign. It was not a CRM implementation. It was not a "digital transformation." It was one AI phone agent trained on his catering menu, his private event rules, his three locations' calendars, and his pricing structure.
The agent picks up every call the humans can't. It greets the caller with his business name. If the caller is asking about reservations for dinner, it routes them to the right location's reservation line. If the caller is asking about catering, it takes the date, headcount, dietary restrictions, and budget, and books a 15-minute follow-up call on his calendar for the next morning.
Setup cost: $4,800. Monthly operation: $950.
Month 1 was intentionally slow. We ran the agent in "shadow mode" for the first two weeks — it answered calls but routed everything to a human for review. This is standard. It lets the operator confirm the agent isn't saying anything weird before it's booking real business. Week 3 we flipped it live.
Payback on setup: day 38.
This is the part that surprises most operators. The recovery line didn't flatten at $5K/month. It compounded, because two things kept happening.
First: the agent got better. Every month we'd review the 5% of calls that went sideways — usually callers asking about something very specific that wasn't in the initial training (private room minimums, off-menu requests, holiday-schedule exceptions) — and we'd teach it. By month 6, the "needs human escalation" rate was down to 3% from 14%.
Second: he started using the recovered hours. Fourteen hours a week of not answering the phone translated, for him, into reopening a weekend brunch service at his flagship location that he'd killed during COVID because he couldn't staff it. The brunch service alone added another $28K/month in net revenue within four months.
Year-end totals:
Net-of-cost recovery: roughly $164,000. Return on Calibrate spend: roughly 11x.
The numbers are the easy part. What the numbers don't capture is what it felt like for him to stop being the backup for the phone. His wife noticed first. His kitchen manager noticed second. By month 4 he was home for dinner every weeknight for the first time in a decade. That's not on a P&L. But that's the thing operators actually mean when they say they want to "run the business instead of being run by it."
The math is the excuse. The quality-of-life change is the actual reason.
He's our most vocal client, off-the-record. The day he decides to go on the record, this page gets his name and his three locations. Until then, every number in this post is exact and sourced from his own phone records, his own POS, and a spreadsheet we reconcile together quarterly.
If you want to know what the equivalent math looks like for your business, the free AI Calibration Call calculates it. Real numbers. No obligation. Written recommendation in just days.