An honest teardown of the first six months of production calls across every industry we serve. No cherry-picked highlights. No marketing spin. Just what happened when real customers called real businesses and an AI answered.
The setup
We reviewed 2,400 calls across HVAC, plumbing, dental, legal, auto repair, and salon businesses. Every call was transcribed and scored on five criteria: did the AI understand the request, did it capture the right information, did it book the appointment correctly, did the caller seem satisfied, and would the call have converted to a job.
What the AI gets right
Speed. Average answer time: 4.2 seconds. No caller waited more than 6 seconds. Compare that to the industry average of 42 hours for a callback. This alone is the biggest factor in customer satisfaction.
Basic intake. Name, phone number, address, service needed, urgency level — the AI captures these correctly 96% of the time. This is the bread and butter, and it works.
Emergency qualification. When a caller says "my AC is out and I have a newborn" or "there's water coming through the ceiling," the AI correctly flags it as urgent and escalates. It catches urgency cues 93% of the time.
Appointment booking. When availability is straightforward, the AI books correctly 91% of the time. "I need someone tomorrow morning" gets handled cleanly.
Tone. This surprised us. In post-call surveys, 88% of callers said the experience was "good" or "great." Several didn't realize they'd spoken to AI until told afterward.
Where it still breaks
Complex scheduling. "I need someone between 2 and 4 on either Thursday or Friday, but not if it's raining because the work is outside." The AI handles this about 70% of the time. Multi-conditional scheduling is its weakest area.
Heavy accents and background noise. Job sites, busy restaurants, older callers with thick regional accents — accuracy drops to about 82% in these conditions. It usually recovers by asking for clarification, but some callers get frustrated by being asked to repeat.
Pricing questions. "How much does it cost to replace a compressor?" The AI is trained not to quote specific prices (because they vary), but some callers want a number and get annoyed by "it depends on your specific system."
Emotional callers. Someone whose basement just flooded doesn't want to calmly provide their address. They want reassurance. The AI provides it, but not as naturally as a human who can truly empathize in the moment.
The honest summary
The AI handles 89% of calls as well or better than the average front-desk employee. It handles the other 11% adequately but not perfectly. Zero calls were handled badly enough to lose the customer — the worst outcome was a caller who had to repeat their address once.
Is it perfect? No. Is it better than voicemail? By every possible measure, yes. The gap between "AI answered in 4 seconds" and "nobody answered for 42 hours" is the gap between winning the job and losing it.
We're improving the weak spots every month. But even today, the AI is catching hundreds of calls that would have gone to voicemail — and that's revenue these businesses would never have seen.