How do you choose an AI receptionist?
Judge it on four things, in this order: whether it can hand a real customer to a real human cleanly, whether it writes what it learned into the system you already use, what happens on the calls it cannot handle, and what the bill looks like in a busy month rather than an average one. Voice quality is the thing everyone demos and the thing that matters least — it is now good across the board. The failures happen at the seams.
Here is the checklist. Take it into the demo.
Before the demo: know your own numbers
You cannot evaluate a quote without these four figures. Most owners can pull them in ten minutes.
- How many calls do you get in a month? Your phone system or mobile log has this.
- How many go unanswered? This is the number the AI is being hired to fix.
- What proportion arrive outside business hours?
- What is a booked job worth to you on average? This turns the whole discussion from a cost into a return.
If you do not know your missed-call number, start there — how to stop missing customer calls walks through it.
On the demo call: 6 things to test yourself
Do not watch a scripted demo. Ask for the phone number and call it yourself, from your own phone.
1. Interrupt it mid-sentence. Real callers talk over the answer. A system that has to finish its sentence before it hears you will frustrate customers within a week.
2. Give it a messy answer. "Uh, I think it's the... hang on... it's the thing under the sink that's leaking." That is what an actual call sounds like. Scripted demos never test it.
3. Ask for something it cannot do. Then watch the escalation. Does it take a message, offer a callback, transfer to a human, or dead-end the caller? This is the single most revealing test on the list.
4. Ask it to book an appointment. Then check your calendar. Did the event appear, in the right slot, with the caller's details attached?
5. Call from a noisy place. A truck cab, a job site, a busy street. Your customers do.
6. Ask it a question about your business that it should not know. You are checking whether it invents an answer. A system that confidently quotes a price you never gave it will do that to your customers.
Pricing: 4 questions that change the number
7. Per minute, per call, or flat rate? Per-minute billing rewards a system that ends calls quickly, which is not always what you want. Per-call billing can charge you for wrong numbers and robocalls. Know which behaviour your pricing model is paying for.
8. What counts as a billable call? Spam calls, hang-ups in the first three seconds, and your own staff calling in — are those billed? Over a year, this is not a small number.
9. What happens when you exceed the plan? Overage rates are where a predictable bill becomes an unpredictable one. Get the number, not the reassurance.
10. What is the contract term, and how do you get your number back? If you port a number in, confirm in writing how you port it out. This is the question that quietly determines how free you are to leave.
Our own pricing is published on the pricing page — and any vendor that will not publish theirs is telling you something.
Integration: 4 questions that decide whether it works
11. Where does the information go? A receptionist that answers well but leaves the details in a dashboard you never open has moved your problem, not solved it. Name the destination: your CRM, your calendar, a text to your phone, an email.
12. Can it read your calendar's real availability? Booking into a slot you are already committed to is worse than not booking at all.
13. What does a human on your team see afterwards? Full transcript, a summary, or a notification? Whoever calls the customer back needs the context without listening to a recording.
14. Who maintains the knowledge it works from? Prices, service areas, hours and policies change. If updating that requires a support ticket and a three-day wait, you will stop updating it — and then it will be wrong.
The three answers that should end the conversation
Be blunt about these:
- "It never needs to transfer to a human." Every system needs an escape hatch. A vendor claiming otherwise either has not run enough real calls or is not being straight with you.
- "You can't see the transcripts." They are your calls and your customers.
- "We can't tell you the overage rate." They can. They would just rather not.
AI receptionist, answering service, or voicemail?
Worth being clear about what you are actually replacing:
| Voicemail | Answering service | AI receptionist | |
|---|---|---|---|
| Answers every call | No | Usually | Yes |
| Available 24/7 | n/a | Depends on plan | Yes |
| Knows your prices and services | No | Limited script | Yes, if maintained |
| Books into your calendar | No | Sometimes | Yes, with integration |
| Cost model | Free | Per minute, usually higher | Varies — check questions 7–9 |
The full comparison is in AI receptionist vs answering service.
The 30-day test that settles it
Do not evaluate on the demo. Run it on real calls for a month, then measure three things:
- Answer rate — what percentage of calls were handled without going to voicemail?
- Booking rate — how many calls became a scheduled job?
- Escalation quality — of the calls it could not handle, how many reached a human who actually followed up?
If answer rate goes up but booking rate does not, the problem is the handoff, not the AI. That is fixable, and it is the most common thing to find in month one.
Frequently asked questions
What should I look for in an AI receptionist?
Four things, in order: clean escalation to a human when it cannot help, integration that writes call details into the CRM or calendar you already use, sensible handling of the calls it cannot complete, and pricing you can predict in a busy month. Test all four yourself by calling the demo number from your own phone rather than watching a scripted demo.
How is an AI receptionist priced?
Usually per minute, per call, or a flat monthly rate. Each rewards different behaviour, so ask specifically what counts as a billable call — spam calls, immediate hang-ups and internal calls are the ones that quietly inflate a bill — and what the overage rate is once you exceed the plan.
Can an AI receptionist transfer calls to a real person?
It should, and any vendor claiming its system never needs to is worth a second look. Test the escalation path on the demo: ask for something outside its scope and see whether it transfers, takes a message, offers a callback, or leaves the caller stranded.
Will an AI receptionist book appointments in my calendar?
Only if it is integrated with your calendar and can read real availability. Test it during the demo by booking an appointment and then checking your own calendar for the event, the correct time slot, and the caller's details.
How long should I test an AI receptionist before deciding?
Run it on real calls for about 30 days and measure answer rate, booking rate, and how well escalated calls were followed up. A demo tells you how the system behaves on a good call; a month tells you how it behaves on all of them.