AI answering services for small business: the receptionist math.

Answering the phone is a staffing problem before it’s a technology problem. The person who can quote a rewire or calm a nervous patient is usually the same person on a ladder, in an exam room, or driving between jobs. So the call rings out, voicemail picks up, and voicemail is where leads go to die.

AI answering services promise a way out, and the comparison articles will happily rank twelve of them by feature count. What none of those lists answer is the sizing question underneath: how much phone coverage does your business actually need to buy, and in which unit? Calls, minutes, or a flat subscription. The unit decides the bill, and the right unit falls out of numbers you already have.

What you’re actually buying

Strip the branding and every product in this category does some slice of five jobs: pick up instantly, say true things about your business (your hours, your service area), capture who called and why, book the next step, and hand off to a human when the call is beyond the script. The first four are largely solved. The fifth is the quality bar.

Anyone’s demo handles “what are your hours.” The caller with a half-finished job and a complaint is what separates products, because that call mishandled costs more than a call missed. It’s the same guardrail problem we walk through in wiring an AI assistant into your stack: the value lives in what the system does when it’s out of its depth.

The three ways they charge

Every vendor bills in one of three units, and the unit tells you who the product was built for.

Per call: the answering-service model

This is the pricing of the human answering-service industry with AI grafted on. Smith.ai, the best-known name here, starts at $300 a month for 30 calls. That’s ten dollars a call, overages at $11.50, and the useful extras metered on top: booking an appointment adds $1.50 a call, SMS notifications 50 cents, call recording a quarter, all at the time of writing. Predictable, mature, expensive per unit. The model makes sense when volume is low and each call is worth real money.

Per minute: the voice-AI model

Platforms that ship a voice agent meter usage instead. HighLevel’s Voice AI, the one we deploy, lists usage charges on top of its $97-to-$497 base plans; the published rate floats around 13 cents a minute, which prices a typical three-minute call near forty cents. Check the current AI pricing page before you budget, because these meters move. The catch is that you’re buying the platform, not just the receptionist. This pricing only exists inside a system that is also your CRM and calendar.

Flat: the add-on model

HighLevel also sells its whole AI bundle (voice agent, chat, review responses) as an “AI Employee” add-on: $97 a month per account for the unlimited tier, $50 for a lighter one. Flat pricing inverts the math. The more calls you feed it, the cheaper each one gets.

Read the table as arithmetic, not a verdict; the columns aren’t equivalent products. Smith.ai’s price includes a human escalation tier the pure AI options don’t, and the HighLevel numbers assume you want the platform underneath. But the arithmetic is the part the listicles skip, and a 25x spread per call is not a detail.

Price the service in the unit you actually lose: booked jobs, not minutes.

Start one tier below: missed-call text-back

Here’s the stance the vendor content won’t take: most local businesses shouldn’t start with an AI receptionist at all. Start with missed-call text-back, the workflow that fires a text the moment a call rings out. “Sorry we missed you. This is Cursive Media — how can we help?” The caller who was about to dial your competitor is now in a conversation instead.

It costs nearly nothing. It ships as a stock workflow in CRM platforms and takes minutes to turn on in HighLevel; the marginal cost is SMS fees. It’s the first thing we wire in a local business automation build.

Its limits are honest ones. A caller who won’t text (plenty of older customers, anyone mid-emergency) gets no rescue, and a burst pipe dials the next number in the map pack rather than wait for your reply. Text-back is the floor of phone coverage, not the ceiling. But it’s a floor that costs pennies, and its recovered-lead count tells you whether the ceiling is worth paying for.

When the full receptionist earns its price

The upgrade case is concrete. An appointment-dense business, a med spa or a dental office where a booked call is direct revenue, converts answered calls to money at a rate that makes even ten dollars a call cheap. Real after-hours volume is the same story: if a third of your calls land outside business hours, an agent that books while you sleep is working a shift no employee wants. And if you already pay a front desk mostly to answer phones, the comparison isn’t $300 versus zero. It’s $300 versus a salary.

Flip those conditions and the case collapses. An owner-operator who answers 80% of calls personally, with a dozen missed calls a month, buys almost nothing that text-back doesn’t already recover.

Go in knowing the failure modes, because every product here has them. Multi-part requests (“reschedule Tuesday, and is the estimate ready?”) still trip agents. Mispronounced names erode trust fast. Some callers hang up the moment they clock a bot. None of that shows up in a demo; it shows up in recordings.

How to trial one without regretting it

Run the trial like an interview, not a demo. Call your own number and be a hard customer: reschedule a fake appointment and add a complaint mid-call. Then read the transcripts. Any serious product records and transcribes, and a vendor who can’t show you a transcript of a difficult call has answered your question. Check where the data lands, because a captured lead that ends life in an email inbox instead of your CRM is a lead you lose twice. Confirm escalation actually rings a phone you control, and that the calendar it books into is the calendar your team really runs.

That last check matters more than the vendor choice. An answering service that books ghosts into an unused calendar is worse than voicemail, because it fails while looking successful.

The wiring is where we come in. Cursive Media builds phone coverage as part of AI integration: text-back first, then a voice agent where the math clears, with everything landing in one system alongside your calendar and follow-up. Book a 20-minute strategy call and bring last month’s missed-call count. We can usually do the receptionist math on the call itself.