Most writing about AI for the trades is written by people who have never watched a call come in at 9pm on a Saturday. We built an AI receptionist for plumbing companies, put it on real customer calls, and learned a set of things that are not in anyone's brochure. This is that list.
Start with the only number that matters
Roughly one in four. Invoca sells call tracking so read them with that in mind — but this is platform data from real calls, not a survey, and it matches what every owner in this trade already suspects.
The reason that number is so much worse than it looks: a person with water coming through their ceiling does not leave a voicemail. They hang up and dial the next name. In a business where a single call can be worth several thousand dollars, a 27% miss rate is not an admin problem, it is your largest unmeasured expense.
In most trades businesses, the phone is not a support channel. It is the sales channel — and it closes at five.
What earned its keep
Answering at all, at every hour. Unglamorous and by far the biggest win. Nights, weekends, and — the one people forget — the moments when the office is already on another call. Two people ringing at once is invisible in your records: the second one never appears anywhere, because nothing logged them.
Triage before booking.The valuable judgment is not "what service do you need," it is does this ring somebody right now.A burst pipe and a dripping tap are the same words in a menu and completely different events. Getting that split right, in the owner's own definition of an emergency, is where a system stops feeling like a robot.
Answering in the caller's language. Ours worked in English and Spanish, picking it up from the caller rather than making them choose from a menu. In a lot of American service markets this is not a nice-to-have, it is a share of the market that currently reaches voicemail.
Writing everything down. Underrated. Once every call is logged and transcribed, you can finally answer questions you could never answer before: how many calls actually come in, what people ask, which of those became money.
What didn't
Sounding impressive. Nobody calling a plumber is evaluating your technology. They want to know someone is coming and when. Effort spent making an agent sound clever is effort not spent making it route correctly.
Automating the parts with judgment in them. Pricing a non-standard job, promising a window you cannot staff, deciding whether to send someone out at 2am — these should reach a human. Building guardrails that escalate early is not a limitation, it is the difference between a system a business keeps and one it quietly turns off.
The model, honestly. The part that answers the phone was the least difficult component. What was hard was everything around it: the telephony, what happens when a transfer fails, what the system does at 2am when a booking API is down. We wrote that up properly in the build story — it is the most useful thing we can hand someone evaluating this.
How to tell if it is worth it for you
Three numbers, and you can get all of them this week without buying anything:
1. How many calls do you actually miss? Your phone provider has this. Almost every owner guesses low — usually because the guess is based on voicemails, and most missed callers never leave one.
2. What share of answered calls become work? Roughly is fine. This is the rate a rescued call would convert at.
3. What is a job worth to you in margin? Margin, not revenue — the number that survives materials and labour.
Multiply, then discount hard: assume only half of the missed callers were real, reachable demand. If what is left is still meaningfully larger than the cost of building the thing, you have your answer. If it is not, do not build it — and be suspicious of anyone who tells you otherwise.
Whether the same holds for your trade
The pattern generalises further than plumbing, and the test above is the same everywhere: find where demand arrives, find where it dies, price the gap.
In home services and the trades the leak is nearly always the phone and the emergency split. In appointment businesses — clinics, salons, studios — it is the calendar, and no-shows usually beat missed calls. In real estate it is speed to lead, because the first responder generally wins the client and inquiries arrive while everyone is out showing houses.
Different leaks, one method. If you would rather not do the counting yourself, that is what the first call is for — it is free, and the written report is yours to keep either way. There is a full example of one here, or you can start one for your business.