An AI receptionist is software that answers a business's calls, asks qualifying questions, books appointments, and hands off to a person when a conversation goes past its script. Disclosure: Mirastart builds them, so read this like a contractor's guide to hiring a contractor. The category is growing fast and its marketing has outrun plain explanations of where these systems fail.
First: three situations where you shouldn't buy one yet
Three situations make an AI receptionist the wrong purchase this quarter: nobody has listened to a week of real calls yet, call volume is low enough that voicemail-to-text would do the job, or there is no reliable person to escalate to. Each one is cheaper to fix than to automate around.
- You don't know what your team actually says on the calls you're trying to automate. If nobody has listened to a week of real calls and written down the common questions, the answers, and where a human takes over, there's nothing yet to build the system around - and a vendor who doesn't ask for this first is guessing.
- Your call volume is genuinely low. A handful of calls a day at odd hours might be better solved by a cheaper answering service or a well-set-up voicemail-to-text flow than by standing up and paying for a full AI system.
- You can't commit to a real escalation path. An AI receptionist that has nowhere reliable to hand off an angry caller or a genuine emergency isn't saving you a problem - it's delaying one and making it worse.
What "AI receptionist" actually covers
The label covers three different products. Some tools only transcribe a voicemail and text it to you. Some answer the call, ask a few qualifying questions, and text a lead notification. Fewer book directly into your calendar, confirm the appointment, and hand a live call to a person. Ask which one you're being quoted.
The differences matter more than any feature list. Ask a vendor to demonstrate the exact call flow on your actual services, not a generic demo script, and listen for where the script runs out. The gap between the pitch and the product usually shows up in the first minute.
Why do businesses buy one?
Businesses buy an AI receptionist to catch the calls a staffed desk can't reach: after hours, during a rush, and on the third line ringing while two are already in use. ServiceTitan's June 2022 platform data put shops with fewer than five technicians at a 24% call booking rate and shops with 25 or more at 59%.
The same report shows the rate moving by hour: large businesses ran 61% at their best hour and 21% after 6 p.m., small businesses 26% down to 9%. "My gosh, the impact and the importance of capturing this thing on the front end can lead to huge results on the back end," says Chris Hunter, founder of Go Time Success Group and a Principal Industry Advisor at ServiceTitan.
Whether the caller is told
A caller should be told they're talking to a system, plainly, near the start of the call. It costs nothing, it's the position that ages well as state disclosure rules for AI voice and chat keep tightening, and it's simply what an honest business does.
California's bot-disclosure statute, Business and Professions Code § 17941, is one example already on the books. A system built to sound indistinguishable from a person on purpose is optimizing for the wrong thing: the goal is a caller who gets a fast, correct answer and knows exactly what they're talking to.
The ownership questions
Six questions decide whether a contract is safe to sign: who owns the phone number, who owns the call data, where a booked appointment lives, how it's priced, what the exit costs, and what the system does when it can't understand a caller. The table gives the good answer and the warning sign for each.
| Ask | A good answer | A warning sign |
|---|---|---|
| Who owns the phone number? | You do, and it's portable if you leave | It's provisioned on our platform and doesn't transfer |
| Who owns the call data and transcripts? | You do, exportable on request | Vague, or 'we retain that' |
| Where does a booked appointment actually live? | Your calendar or CRM, as the source of truth | A dashboard only their platform can read |
| How is it priced? | A clear flat fee or a stated per-minute rate up front | Pricing revealed only after a sales call |
| What's the contract length and the exit? | Month-to-month, or a stated term with a stated penalty | Locked in for a year, discussed only after you're sold |
| What happens when the system doesn't understand the caller? | A defined, fast handoff to a person | It keeps guessing or the call just ends |
Recording and consent aren't optional reading
North Carolina is a one-party consent state for recording calls (N.C. Gen. Stat. § 15A-287), so your business, as a party to the call, can legally record without announcing it. That's the legal floor, not the bar to build to, and callers in all-party states like California sit on the other side of it.
A business serving customers beyond North Carolina should default to a brief recording notice regardless of what the home state technically allows. Ask any vendor plainly how their system handles consent, and don't accept "that's on you" as the full answer: if they're building the call flow, they're part of that decision.
Pricing models, honestly
Three pricing structures cover most of the market: a flat monthly fee, a per-minute or per-call rate, and a per-booked-appointment fee. Each one is defensible and each one has a failure mode, so the right structure is whichever matches how your call volume actually behaves across a year.
| Structure | What it suits | Where it breaks | What to ask |
|---|---|---|---|
| Flat monthly fee | Steady, predictable call volume | Thin value in a quiet month, and renegotiation pressure in a loud one | What happens in my busiest month? |
| Per-minute or per-call | Low or seasonal volume | Costs the most in exactly the month you needed the help | Is there a cap, and what counts as a billable minute? |
| Per-booked-appointment | Buyers who want cost tied to outcome | The definition of a booking can drift | What counts as a booking, and are cancellations refunded? |
No structure is universally right. Price each quote against your own call pattern for the last twelve months, peak month included, before you compare the headline numbers.
The integration question
An AI receptionist that doesn't talk to your calendar and your CRM turns into one more inbox someone has to check by hand, which defeats the point. Before signing, confirm the system writes directly into the tools you already run rather than into a separate dashboard nobody opens after week one. If it can't integrate cleanly, that should be flagged during scoping - not discovered a month in, after the team has quietly stopped trusting it.
What this looks like when it runs
Booking, missed-call follow-up, review requests and reporting run as one system in production for Charlotte businesses. The call is answered or captured, the appointment writes into the calendar you already use as the source of truth, the follow-up text goes out on a timer, and the monthly report counts booked jobs.
Sources
- N.C. Gen. Stat. § 15A-287, Interception and disclosure of wire, oral, or electronic communications prohibited - North Carolina's one-party rule: interception is an offense only when done "without the consent of at least one party to the communication."
- California Penal Code § 632 - An all-party-consent state: recording a confidential communication requires the consent of all parties.
- California Business and Professions Code § 17941 (bot disclosure) - A state bot-disclosure statute - one example of the AI-interaction disclosure rules tightening across states.
- Federal Communications Commission, Declaratory Ruling FCC 24-17, CG Docket No. 23-362 (released February 8, 2024) - AI-generated voices count as "artificial" under the TCPA; any outbound calling the system does needs prior express consent.
- ServiceTitan, Data Report: Average Call Booking Rates - June 2022 platform data from more than 3,000 US and Canadian trade businesses: shops with fewer than five techs booked 24% of calls, shops with 25 or more techs 59%; by time of day, large shops ran 61% at peak and 21% after 6 p.m., small shops 26% to 9%. The Chris Hunter quotation is from the same report.