A 24/7 voice receptionist and a review chain that knew which reviews to chase.
Two problems, one stack: missed calls after-hours were leaving five-figure revenue on the floor every month, and post-visit reviews were inconsistent — when a happy client got the request, half the time they didn't follow through. We installed a voice AI front door and a review-capture chain that routes intelligently based on sentiment.
The problem
This group runs four locations. Their clients don't think about office hours — they call when they get off work. After 6pm, the front desk was rolling to voicemail, and roughly a third of those callers never called back. That's a quiet, unmeasurable bleed: tens of thousands of dollars in missed bookings every quarter, gone before anyone could even see it.
The review situation was the second drag. Their ops manager would text past clients ad-hoc asking for Google reviews, and the conversion was bad. Worse, when a client had a so-so visit, the request still went out — and they got the bad review they could have prevented if they'd known to call first.
The system we built
Two chains, one stack. The first: a 24/7 voice receptionist (VAPI primary, ElevenLabs voice) that answers, qualifies, books into the right location's calendar, and SMSes confirmation. After-hours calls get the AI; in-hours calls get the front desk first and the AI as fallback.
The second: a post-visit review capture chain. 24 hours after a visit, the client gets a friendly text asking how the visit went. If they reply positive, the chain sends them straight to Google. If they reply neutral or negative, ops gets pinged within 60 seconds — long before the client thinks to leave a public review.
Voice receptionist flow
Review-capture chain
What was harder than it looked
The voice AI needed to handle five different service categories with different qualifying questions, across four locations with different hours and different specialists. We built a routing layer in front of VAPI that picks the right config based on caller intent + location detection. We also added a Spanish-language handoff (SVC-08) after the first month — about 14% of after-hours calls were Spanish-first.
For reviews, the trick was the sentiment classifier. We didn't want false negatives sending unhappy clients to Google. We tuned the threshold conservatively, and we added a HITL gate on borderline cases — ops sees them in a Slack channel and approves with one click.
"We thought we were doing fine on bookings. Then we turned this on, and it turns out we were missing a third of our pipeline. After-hours is real business. We just couldn't hear it ringing."
— Operations director (name withheld)
What we measured
- 38% of bookings now happen after-hours — calls the front desk would have lost. That's the headline number, but the second-order effect is what got the owner's attention: the ROI hit positive in 19 days.
- +62 net new Google reviews per quarter, averaged across locations. The 5-star ratio went up because unhappy clients now go to ops, not Google.
- 94% voice-AI booking-success rate. The 6% that fall through hit a live human within the next business hour.
- 11 days from audit to first live call answered by AI.
What's running today
Both chains run in production across all four locations. We added a no-show one-click rebook (ONB-04) atom in week 6 of tuning — about 9% of voice-AI bookings were no-shows in the early data, and the rebook chain claws roughly half of them back. Voice AI handles roughly 280 calls a week.