Most pages selling an AI receptionist for small business list everything it can do. This one starts with what ours refuses to do, because after testing a live agent against callers who pushed back, the refusals turned out to matter more than the features.
It answers every call, 24 hours a day. It books, reschedules and cancels appointments against your live schedule. It answers questions from your own uploaded documents. And it will tell a caller "I don't know" rather than guess.
An AI receptionist answers your business phone, holds a normal spoken conversation with the caller, and completes a task — usually booking an appointment or capturing the caller's details for you. Unlike an answering service, it doesn't take a message for someone else to action later; it does the thing while the caller is still on the line.
For a small business that means three specific jobs. It picks up on the first ring when you're with a customer, on a job, or asleep. It checks what's actually free in your schedule and books into it. And it logs the call with a transcript and a factual summary, so you can see what happened without listening back.
Because a wrong number quoted on the phone is a problem you inherit, not the agent. Ours will explain that pricing depends on the job and offer to book an assessment, but it will not invent a figure — even when the caller pushes.
This is the part we got wrong first, and it's worth being specific about how.
When we tested our own agent on a tree service configuration, a caller asked what a removal would cost. The agent didn't invent a price — good — but it also didn't answer the question. It launched into its full qualifying checklist, five questions deep, before acknowledging what the caller had actually asked. On the recording it reads as stalling. A homeowner who wants a ballpark and gets an interrogation hangs up.
So we changed the instruction: on a pricing question, ask at most one clarifying detail, then give an honest explanation that the price depends on the job and offer the assessment. We retested it twice — once on the straight question, once on a caller pushing with "come on, just give me a ballpark." No invented number either time, and the same explanation both times.
That's the trade being made. You lose the caller who wanted a number on the spot. You don't get the customer who was quoted $400 on the phone and then handed a $900 invoice.
If you'd rather it did quote — a fixed call-out fee, a flat rate for a standard job — you can tell it exactly what to say, and then it will say exactly that. What it won't do is fill the gap on its own.
It says it doesn't know, and offers to take your details so someone can call you back. It answers from the FAQ and policy documents you upload and nothing else, so it has no way to improvise an answer about your insurance list, your service area, or your cancellation policy.
This sounds like a small thing until you consider what the alternative does to you. An agent that guesses at "do you take Delta Dental?" creates a patient who arrives expecting coverage they don't have. The honest failure is cheaper than the confident wrong answer, every time.
No — and this is the refusal we'd most want you to test yourself. If a transfer isn't available, the agent says so and takes the caller's name, number and reason so your team can call back. It doesn't say "putting you through now" into silence.
We found this one the hard way too. Testing an emergency scenario — a caller reporting a tree touching a power line — the escalation trigger fired correctly, but the transfer itself didn't complete in our test environment. What mattered was what happened next: the agent didn't pretend the transfer had worked. It gave the caller the safety instruction to contact the utility company, then offered to take callback details without being asked to.
Be careful how you read claims about this across the category, including ours. "Transfers to your team with a summary in hand" is only true where live transfer has actually been configured and tested for that specific deployment. Until it has been, the honest description is that it takes the details and your team calls back. Ask any vendor to demonstrate a completed transfer rather than describe one.
Yes, and in order, one at a time. For an urgent job it works through the situation before offering a slot — what's happened, what it's affecting, whether anyone is in danger, then access and timing.
This was the biggest defect we found in our own testing, and the fix was structural rather than cosmetic. The original agent bundled questions together, or skipped straight to booking when the caller sounded ready. That produces an appointment with half the information, which means someone on your team has to ring back and ask anyway — the exact work you bought the agent to remove.
Rebuilding it as a numbered per-category checklist, one question per turn, fixed it. On retest of a storm scenario the agent worked through the sequence without bundling, held the sequence even when the caller pushed on urgency, read the name, phone and address back accurately, and — this is the part that mattered — used the caller's stated urgency to pick between two different appointment types rather than just recording it.
Safety still overrides the checklist. A caller in immediate danger gets the instruction first, not question three.
An answering service takes a message and sends it to you. An AI receptionist completes the booking while the caller is on the phone, so there's nothing left in your inbox to action.
The distinction matters most for the calls you're actually losing. Someone calling three businesses about a broken furnace isn't waiting for a callback — they're booking with whoever can give them a time. A message that reaches you at 6pm is a record of a customer you already lost.
Worth being fair about the other direction, though: a human answering service handles genuinely unusual calls better than any agent does, and if your calls are mostly complex, sensitive, or emotionally difficult, a person is the right answer. Services like Smith.ai combine AI agents with live North American receptionists precisely because some calls need a human. An AI receptionist earns its place on the repetitive ones: hours, location, availability, booking, rescheduling.
It books into Google Calendar and Cal.com. It reads live availability, writes the booking back, and respects appointments made outside the system, so it won't double-book you.
Here's the limit, stated plainly: it does not integrate with practice management systems or field service software. Not Dentrix, not Open Dental, not Jobber, not Housecall Pro, not ServiceTitan. If your schedule lives in one of those and isn't mirrored to Google Calendar, the honest positioning is qualified lead capture with immediate handoff — the agent answers, qualifies, and sends your team the details — not automatic booking.
Several vendors in this category are vague on exactly this point. Ask which calendar, specifically, and ask to see a booking land in it.
Published pricing in this category runs from roughly $49 to $300 a month for self-serve tools, with per-call services costing considerably more. Vox Front AI is priced differently because it's set up for you rather than sold as software you configure yourself.
Vox Front AI's structure is a one-time setup fee, a monthly plan with an agreed amount of call time, and a per-minute rate for anything beyond it. It's month to month, cancel with 30 days' notice, and we give at least 30 days' notice before any price change. You'll know all four numbers — setup fee, monthly price, included minutes, extra-minute rate — before you start.
There are free tiers and free trials, and they're a reasonable way to hear what the technology sounds like. What they generally don't include is calendar booking, answering from your own documents, or anyone configuring it for your business.
That last one is the real difference. A free tool hands you a blank agent and a settings page. Whether it works depends entirely on how much time you spend writing its instructions, testing it against awkward callers, and fixing what you find — which, based on doing exactly that, is not a small job.
That depends on numbers only you have, so here's the arithmetic rather than a claim. Take what an average job or new customer is worth to you, multiply by how many calls you think you miss in a week including evenings and weekends, then multiply by your close rate when you do pick up.
We deliberately don't publish a percentage-uplift figure or a recovered-revenue number, because we have no measured client results yet and an invented one would be worth nothing to you. If your own arithmetic doesn't clear the monthly cost comfortably, it isn't worth it, and no page should talk you past that.
The businesses where it clears easily have three things in common: the phone is how the sale happens, one busy person is answering it, and a missed call goes to a competitor rather than into a queue.
No product can be "HIPAA compliant" as a property of itself — no such certification exists, and any vendor claiming it is telling you something that isn't true. What can be true is that a service is built on HIPAA-eligible vendors with Business Associate Agreements in place, and that has to be confirmed for your specific deployment rather than assumed from a badge.
What we can describe concretely: transcripts are scanned before they're stored, and identifiers like card numbers, Social Security numbers and phone numbers are redacted automatically. If you're a medical or dental practice, ask us about BAAs directly and we'll tell you exactly where things stand rather than pointing at a compliance page.
Most businesses are live in under 10 minutes once the calendar is connected and the FAQ and policy documents are uploaded. That's the technical setup, and it's genuinely quick.
What takes longer, and matters more, is testing. Being live is not the same as being right — our own agent passed a configuration review and still skipped its qualifying questions on real test calls. Before you point your business number at any agent, make a list of the ten calls you actually get, including the awkward ones, and place them yourself.
It keeps answering — calls never get cut off for running over. Extra minutes are billed at the per-minute rate on your plan, and those rates are public on the pricing page rather than something you find out after the fact.
If you're running close to your included minutes most months, it's worth moving up a tier rather than paying the per-minute overage rate indefinitely — the numbers for each plan are laid out on the pricing page so you can do that math yourself.
Vox Front AI works with service businesses across the United States. It's a remote, built-for-you deployment — there's no office to visit. Setup happens over a call, and the agent runs on your existing business phone number.
The only proof worth anything in this category is a call you make yourself. We'll load a demo agent with your real hours, services and policies, and you can try to break it — push it on price, ask it something it shouldn't know, insist on a human.
Book a demoAbout the author
Kainat is the founder of Vox Front AI and configures and tests every deployment herself. The testing described on this page — the qualification defect, the pricing regression, the power-line escalation — comes from her own test call transcripts, not from vendor documentation. Read what broke when we tested this ourselves.