What does a bad AI receptionist cost you in lost jobs and bad reviews?

The call log says it was answered. The calendar says Thursday, noon to four. Your tech pulls up just after noon, and the house is the wrong one: the receptionist heard the street number wrong, read it back fast, and the caller said yes without really listening. Across town, the homeowner who actually called is watching the driveway.

Here's the short answer. A bad AI receptionist can send you three bills where a missed call sends one. It loses jobs while looking like it answered. It puts bookings on your board that send a tech to the wrong house or for work you don't do. And it makes promises, like "you're booked," that turn a caller into a complaint.

No published study puts a dollar figure on any of the three, so the honest number comes from your own books. Price each failure at what it actually costs you: the job, the tech's hour, the customer. Check the promises first, because the failures a caller believes are the expensive ones.

Full disclosure: Willison sells a managed AI receptionist, so we have a side here. Some of the failures below can happen on ours too, and the last section says which ones and what's built against them. There's also a section on when a bad receptionist is worse than plain voicemail.

Three bills a bad call can send you

A missed call sends one bill: the job that went to whoever answered. A receptionist that answers badly can send that one and two more. Here's which failure lands where, and who finds out first.

• The lost job, found by nobody. It never offers a time, takes a message that never reaches you, or loses the caller in a loop of questions. Your log shows an answered call.
• The wasted trip, found by your tech at the curb. It mishears the address, files the job under the wrong service, books work you don't do, or puts two customers in one slot.
• The angry customer, found by you, later. It tells a caller they're booked when nothing's on the board, gives an arrival time nobody set, makes up a price you never gave it, or claims a licence you don't hold.

The first bill is the hardest to see. The third can cost the most per call, because the job goes and the customer can keep talking about it.

What does it cost when it sounds fine and books nothing?

This bill doesn't show up anywhere. The call was answered, the caller hung up polite, and there's no missed call to return.

The nearest yardstick for what a decent answer should do is Invoca's 2026 home services benchmarks. On calls answered by a person, 45% of the leads converted on the call. And on 55% of the home services calls Invoca scored, nobody asked the caller to buy or book.

Those are people answering, at businesses that track calls with Invoca, not AI receptionists and not your shop. They show that "answered" and "booked" are different results for anybody.

A receptionist that answers and never offers a time leaves the caller about where voicemail would. Invoca's platform data found fewer than 3% of callers sent to voicemail leave a message. A caller told "someone will call you back" is in the same spot, with the next number on Google one tap away. Voicemail at least leaves a missed call in your log.

The same bill comes from a message taken well and then lost, in an inbox nobody checks after five. Pricing it takes two of your own numbers: how many of these calls you had, and what that job is worth to you.

What does a wasted truck roll cost?

A receptionist doesn't send trucks. It puts a booking on your board, and the bill lands when nobody checks that booking before the tech rolls: the first job of the morning, a one-truck shop, a Saturday.

The bookings that cost a trip:

• An address misheard, or a unit number dropped.
• A job filed under the wrong service, so the wrong tech shows up, or the right one without the parts.
• Work you don't do, like a water heater at an HVAC-only shop.
• Two customers put in the same slot.
• A caller outside your area. Ours captures the address but doesn't enforce your service area, so that one is for whoever reviews the board.

The floor on that bill is the tech's paid time. In BLS OEWS figures for 2025, the median HVACR mechanic earned $29.33 an hour, plumbers and pipefitters $30.67 and the median roofer $26.65. That's wage only, before payroll taxes, the truck, the fuel, and the paying job that hour could have held.

So use your own loaded rate, and add the job that got pushed. And a wrong address you took down isn't a trip charge you can bill the customer for. And if your techs are paid by the job, a dead trip costs them too, which can show up later in who stays.

Which mistakes turn into a bad review?

Picture the caller with the most reason to write a one-star review: the one who took the morning off for a truck that was never coming.

Four promises set that up:

• "You're booked," with nothing written to the calendar.
• An arrival time nobody gave it.
• A price for the job, made up on the spot.
• A licence, insurance or certification you don't have, which is also a claim made in your name.

Add one more: a receptionist that claims to be a person when the caller asks. Someone who finds out later has a reason to doubt everything else it told them.

Notice who gets the review: your company, not the software. There's no published figure on how often AI receptionists cause bad reviews, and we won't invent one. What you can do is check every promise yours makes against the board.

Where do those costs show up in your own books?

Not in the receptionist's dashboard. A count of calls answered can't see any of the three bills. Look where they land, in this order.

1. Your scheduling software. Compare booked jobs per inbound call against the same weeks last year, or weeks with similar weather, since a heat wave or a freeze moves both numbers. Count after-hours calls separately: a receptionist that answers calls that used to ring out adds calls that book less often, which lowers the ratio without anything going wrong. A drop in the daytime ratio is the first bill.
2. The jobs that didn't happen. Wrong addresses, work you don't do, bookings cancelled at the door. Ask your techs to flag every wasted trip for a month, with the reason.
3. The office phone. Callbacks that start with "I was told" or "nobody showed up" are the third bill arriving.
4. Review alerts. Read any new review that mentions the phone, a booking or a no-show.
5. Then the recordings. Pull the calls behind each flag above, plus a handful that ended with no booking and no message, and listen for which failure it was.

Hold whatever answered before to the same standard. Its weeks likely had lost jobs of their own, easy to forget because they left no trace in the log. Compare the two on booked jobs, net of wasted trips and complaints, not on calls answered.

Is a bad AI receptionist worse than voicemail?

It can be. Voicemail sends one bill. Few callers leave a message on it, but it never tells anyone they're booked, never quotes a price, and never puts a wrong address on your board. A receptionist that books some jobs but also makes promises it can't keep sends all three, and the third keeps costing after the call.

So if yours keeps failing on the promises, take it off the line and go back to what you had while it gets fixed. A wrong job type can be fixed as a setting. A "you're booked" with nothing on the calendar isn't, and it shouldn't get a second week on your phone.

The other options send bills too. A live answering service puts a trained person on the line, and a person can still skip the ask, as Invoca's numbers show for people answering generally, or relay a message hours late. Voicemail and a phone left ringing can lose the job outright, a bill that's easy to pay without ever seeing it.

What changes on a restoration call?

On a water loss, the bills come bigger and faster. The call has to capture what the team needs to send the right people: where the water's coming from and whether it's clean or sewage, whether it's stopped, how long ago it started, which rooms and floors are wet, whether the caller owns or rents, and whether a claim's already open. Miss the source, or whether it's still running, and the team shows up planned for a different job.

Speed is the other half. The EPA says to dry water-damaged areas and items within 24 to 48 hours to prevent mold growth, so a water call can't sit as an unread message until morning. And the caller with water on the floor may already be dialing whoever might get there first.

The job at stake is real money. Triple-I, from ISO/Verisk, puts the average homeowners claim for water damage and freezing at $15,400, a weighted average for 2019 to 2023. That's the size of the whole claim, not one restorer's invoice. But it shows how much rides on the call, and a missed or fumbled one can hand the job attached to it to someone else.

Willison doesn't send crews or promise arrival times. On a restoration call it triages the emergency, the source, the water category, how urgent it is, the address and the details, and hands that to your team. That's also why the promise rules matter more here: "someone's on the way" is the one sentence it should never say.

Ranked: ways to answer the phone, by how each is built against the three bills

Ranked by how each option is built against the three bills, best first. Each comes with a test you can run on any vendor, us included.

1. Willison: the best all-around, a managed AI receptionist that books the job and runs the follow-up. It answers every call in seconds, books to your calendar, and is built against the promise bill. It doesn't say "you're booked" until the booking is written, and it reads that confirmation from a fixed line the AI can't reword. It books only the services on your list, states your own fee setup without ever pricing a job, never claims a credential you haven't proved, and says it's a virtual receptionist when asked. It follows up a hesitant caller by text, so a maybe doesn't sit until it's someone else's job. The founder personally reviews every account before it goes live. It can still mishear an address, which is why the last section exists. Test it: ask for a price, ask "are you a real person?", and rattle off an address fast.
2. A well-run live answering service. A trained person on the line, and a person can handle an odd call. It can send the first two bills: a skipped ask, a late message, a misheard address. It's less likely to send the third, because a trained person is far less likely to make up a price. Test it: call at 9pm with a real-sounding job and time how long the message takes to reach you.
3. A DIY AI tool you set up and test yourself. Yours to control. Its guards are only as good as your setup and your testing. Test it: book a job by phone, then check what the caller was told against what landed on the calendar.
4. Voicemail. It makes no promises, so it sends one bill. But it's a big one. Test it: count a week of messages against a week of missed calls.
5. Letting it ring. The same lost-job bill, with nothing left behind but a number.
6. An AI receptionist that makes promises it can't keep. The option most likely to send all three bills. If yours fails the "you're booked" check, it belongs at the bottom until it's fixed.

What our guards don't fix

Guards narrow the bills; they don't erase them. Willison can mishear an address, file a job under the wrong service, get interrupted and restate something wrong, or take a message that doesn't reach you. The read-back, one question per turn and the message to the owner are there because those things can happen.

Some of it no guard catches. A caller who says yes to a wrong address on the read-back has confirmed the mistake for you. The fix sits on your side of the call: a tech who rings the customer before rolling catches it for the price of a minute, and so does a quick look at tomorrow's board by someone who knows the town.

If you want to hear the read-back and the booking for yourself, press play on a real call on willisonhq.com. It's our receptionist answering for a sample HVAC shop, start to finish. For the test calls worth running before you commit to anyone, see how to evaluate an AI receptionist during a free trial.

Frequently asked questions

We haven't found one. Willison doesn't publish a per-call error rate, we haven't found a vendor that does, and review sites don't record what answered the phone, so any number you see is a guess. The number worth having is your own: for a month, flag every review and complaint that mentions a booking or a no-show, and trace each one back to the call it came from.

Your company. They called your number and booked with your business, so a broken promise reads as yours, whatever picked up. Unless they asked, they may not know it was an AI at all. One that answers honestly when asked, as Willison does, at least doesn't add a lie to the complaint.

The promise failures show up within days if you look for them: hold every booking it made against what the caller was told, and have techs flag wasted trips. The lost-job failure takes weeks, because it needs booked jobs measured against a baseline from before the switch. Willison's free trial is seven days of live service starting the day it goes live, with payment details taken at checkout and nothing charged unless you let it run past day seven, so it's enough for the first check but not the second.

That depends on the failure. A wrong job type, a missing service or the wrong hours are settings, so fix each once and test it again with a call. Telling callers they're booked when nothing's written, or inventing prices, is a product failure rather than a setting. Ask the vendor to show you the change, test it yourself, and switch if it fails a second time.