AI vs human answering services: cost and capacity

The handoff is the product, so here's what a good one contains

Whatever answers your phone, you're buying the message. It needs six things before it's worth the interruption:

• The job, in the caller's words. "AC running, air not cold, outdoor unit humming." Not "HVAC issue".
• The address. Street and city. Round Rock and Hutto are twenty minutes apart at 4pm.
• Urgency, with a reason. Indoor temperature. Anyone elderly, an infant, a medical need. A walk-in cooler on a commercial account.
• The callback number, read back. Caller ID is wrong more often than you'd think.
• New or existing customer. Changes who you call first.
• The time the call came in. So you know how long she's been waiting.

Mine does that. It does not book. I'm asked about this on every call: it doesn't book because it doesn't know where your vans are. A 2pm slot in Hutto is a promise. Booking it while both vans are on a changeout in Cedar Park is a promise you'll break, and the customer remembers the broken promise, not the fast answer. Dispatch is a live judgment made with the board in front of you. The AI's job is to make your callback a forty-second call instead of a fishing trip. Catch, qualify, hand off. How that fits an HVAC shop is on the HVAC page.

Cost per call: the maths from the vendors' own pages

These prices are from the vendors' public pages as I read them in September 2026. They'll move. The method won't: take the plan price, divide by the calls it covers, add what the plan leaves out. For minute-billed services, assume a four-minute call, about what a qualifying call takes when the caller is stressed. Say you take 300 calls a month. UK readers: the names on the invoice change, the shape of the maths doesn't.

AnswerConnect sells real people, 24/7, by the minute. Their Growth plan is $395 a month for 300 minutes. At four minutes a call that's 75 calls, or $5.27 each. For your 300 calls you'd need 1,200 minutes, and their 1,200-minute plan is $1,625 a month: $5.42 a call. Every interaction is rounded up to the nearest minute, so three minutes ten seconds bills as four.

Ruby is the premium end. Their pricing page lists 500 receptionist minutes for $1,725 a month, which is $13.80 a call at four minutes. For 300 calls you'd be into their larger tiers: the 1,000-minute plan at $3,350 plus 200 overage minutes at $3.65 is about $4,080, or $13.60 a call.

Smith.ai's human receptionists bill per call, not per minute, and the 2am call costs the same as the 2pm one. Their Pro plan is $2,100 for 300 calls, so $7.00 a call, plus $0.50 a call if you want the message by SMS. Call it $7.50. Calls past 300 are $8.50 each.

The same company's AI receptionist is a different animal. Their Enterprise tier is $500 a month for 300 calls, which they put at $1.67 a call. Same firm, same phone line, roughly a quarter of the human price.

Goodcall bills by unique caller, not by call or minute. Their Growth plan is $129 a month for 250 unique customers, then $0.50 for each one after. Suppose your 300 calls come from 300 different numbers: $129 plus $25 is $154, about $0.51 a call. A caller who hangs up without speaking isn't counted, and the same customer calling three times counts once.

Now mine, and I'll be straight about what I won't say. My builds start at $4,500, fixed scope, fixed price, quoted before I start. The monthly fee is quoted with the build, and I won't put a per-call figure here that pretends otherwise. What the build pays for is a script written from real call transcripts, the handoff text set up the way your dispatcher wants it, a fallback route to your existing line, and me on the phone when something breaks. If you want the cheapest per-call number on this page, Goodcall has it. If you want the message to arrive the way I described above, that's what I build.

Which one, for your shop

• Under about eight calls a day, and you answer most of them. Answer it yourself and call back within the hour. Don't buy anything.
• Seasonal spikes with a quiet shoulder. AI. You're paying for the hour that breaks, not the average day. A human service prices you on the average and queues you on the peak.
• You have a dispatcher or an office manager. Keep them on the phone in hours. AI takes overflow and after-hours, and the handoff text goes to the dispatcher's phone, not yours.
• Your callers need a person to hear them. A property manager with a contract dispute. A customer with a complaint. A human service, or you. Not AI.
• You need the appointment booked on the call. Not mine, by design. A human service with calendar access will do it for a per-call fee, and you'll live with the double-bookings, because they don't know where the vans are either. A CSR with the dispatch board in front of them will do it properly.

What I'm doing about it, and what you can do this week

One of my plumbing clients is replacing its human answering service with my AI receptionist. It goes live in October 2026. I'll publish the first 30 days on /built, including the calls it got wrong, because a comparison written before the data is in should say so. Until then, everything above is vendor pricing and worked examples, not my results.

This week, do one thing. Pull your call log for the hottest week this summer and count three numbers: calls in, calls answered, and calls after 5pm. If the gap between the first two is bigger than your vans can explain, you've found the leak. If you want a second pair of eyes on those three numbers, book a growth call. It's free, thirty minutes, and I'll tell you if the honest answer is to keep your money.