After you buy an AI receptionist, the work is not finished. It starts. Someone should test it on your real calls before your customers hear it, listen to its calls once it is live, fix what it gets wrong and show you every month what your phone did. If the help stops when you sign, you bought a tool, not a front desk.
Who this is for: owners and office managers of home service companies who bought an AI receptionist or voice agent and stopped hearing from the seller, or who are about to buy one. The examples use a heating and air company. The same steps work for any shop that books jobs by phone.
How we know: in October 2026 we read owner reviews of AI receptionist and answering services, and owner forum posts about them. Most reviews of the sellers with the most reviews were good. The one-star reviews kept coming back to two stories, and owner forums added a third.
What do owners say goes wrong after the sale?
- The seller went quiet. Communication was great until they signed. After that, owners could not get a person to fix what was broken.
- Leaving was hard. Owners described cancel requests that were ignored, and charges that kept coming after they canceled or stopped using the service.
- It did not work on their calls. Owners said the AI could not handle the calls their business actually gets.
None of the three is about the sales pitch. All three are about what happens after you sign. A voice agent is software. Someone has to set it up for your business, check it and keep it current. If nobody is named for that job, it lands on you.
What should happen before the AI answers a real customer?
Four things should be done before go-live, and you should see each one.
- Your rules are written down. Your service area and hours. The jobs you book and the ones you pass on. What counts as an emergency at your shop, like no heat on a winter night or water on the floor. What it may say about price. Who gets the call when a person is needed.
- It is tested on your calls. Someone runs your common calls through it: a no-cool call on a Saturday, a caller outside your area, a caller who asks for a person, a price shopper, and a maintenance plan customer who wants to move a tune-up. Every wrong answer gets fixed before go-live.
- You call it yourself. From your own phone, after hours, with your most common emergency.
- The handoffs work. A booked call lands on your dispatch board or calendar. A transfer reaches the right phone. The text to a missed caller says what you want it to say.
What should the first weeks after go-live look like?
Here is what good help looks like.
- A person listens to real calls. Often in the first weeks, then on a set schedule. Not a dashboard summary: the recordings, heard by someone who can change the setup.
- Mistakes get fixed fast. When it answers wrong, the fix should land within days, and you should hear what changed.
- You get a short update. What they heard, what they changed and what they are watching next.
- Bookings get matched. Each job it booked is checked against your dispatch board, so nothing it promised a customer slips through.
What should you get every month after that?
- A plain report. Calls in, calls answered, calls handed to a person, jobs booked and missed callers texted back. If a number is zero, the report says zero. Our guide on how to measure an AI front desk shows which of these numbers prove a result and which only prove activity.
- What changed and why. Every change to its answers, with the call that caused it.
- Updates when your business changes. New hours, a new service area, a price change, a heat wave. The AI should hear about them before your customers do.
What should you own if you leave?
Ask before you sign. A good answer to each of these is "you keep it."
- Your phone number.
- Your call recordings and transcripts.
- Your customer records and bookings.
- The written rules and answers your AI uses.
Then ask how canceling works: how you cancel, how long it takes and when the charges stop.
Which questions should you ask before you sign?
| Ask this | A good answer | A warning sign |
|---|---|---|
| Who do I call when it handles a call wrong? | A named person, and when they will get back to you | A help center link or a ticket form |
| Who tests it before my customers hear it? | They test it on your real call types, then you call it yourself | They say it works out of the box |
| How often does someone listen to my calls? | On a set schedule, and more often in the first weeks | Only when you report a problem |
| What do I get each month? | A short report with calls, bookings and what changed | A login and nothing else |
| What happens to my number and recordings if I leave? | You keep them | They stay with the seller |
| How do I cancel? | A clear answer on how it ends and when the charges stop | You are told to call and discuss it |
What can you check this week if you already have one?
- Pull ten recent calls and listen to them. Mark each one: handled right, handled wrong or handed to a person.
- Write down the three questions it gets wrong most often.
- Call it yourself after hours with your most common emergency.
- Match one week of its bookings against your dispatch board.
- Send your seller the six questions above and see how fast a person answers.
If a person answers fast and fixes what you found, keep going. If not, you know what is missing. For what should happen to the calls nobody picks up, see what should happen when a call goes to voicemail.
How does FlowChainLabs run an AI receptionist?
You do not get a tool to set up. We build the front desk around your trade and your rules, run it, and listen to the calls every month. You hear it take your calls before it goes live. See how the call side works on our AI receptionist page.
Already bought one and stopped hearing from the seller, or about to choose one? Schedule a strategy call. If you have them, bring a few recent call recordings and your booking rules.
