AI Answering Service: What It Does, What It Costs, and When It Is Worth It
We build and operate AI systems for small service businesses, which means we also get to watch them fail. This is what an AI answering service genuinely handles, what it costs against the human alternative, and the situations where we would tell you not to bother.
An AI answering service answers your phone with a voice model rather than a person. It greets the caller, works out why they are calling, captures what you need, and then either books something, transfers the call, or sends you a message. That is the whole product.
The category has grown quickly enough that most of what is written about it is written by people selling it, including, in fairness, us. So the useful thing we can add is not another feature list. It is the parts that only show up after the thing is live.
The statistic everyone quotes, and the one worth using
Almost every page selling this will open by telling you 62% of calls to small businesses go unanswered. It is worth knowing where that number comes from before you make a decision on it.
It traces back to a study by 411 Locals, a lead generation company, which called 85 businesses across 58 industries and found 37.8% were answered by a live person. Eighty-five businesses is a small sample, and the organisation that produced it sells lead generation. That does not make it false. It makes it a marketing survey rather than a measurement, and it has been repeated so widely that it now reads like census data.
The number we would rather work from comes from call analytics rather than a survey. Invoca, which instruments real call traffic for its customers, reported an average missed call rate of around 27% in home services in 2024. Lower than 62%, still high enough to matter, and grounded in calls that actually happened rather than calls placed to test people.
Roughly one call in four goes unanswered at a typical service business, and the rate climbs sharply for solo operators and anyone who spends the day on a job site or in someone's home. That is a real problem. It is not a 62% catastrophe, and you should be suspicious of anyone who needs it to be.
The other figure worth having is timing. A commonly cited breakdown puts around 43% of calls to small businesses outside standard opening hours. That one matches what we see in practice: the calls that go missing are rarely spread evenly through the day. They cluster in the evening, at weekends, and during the two hours a tradesperson spends driving.
Which points at something useful. If your missed calls are concentrated outside hours, you have a coverage problem and almost any form of after-hours answering fixes it. If they are concentrated at 11am on a Tuesday, you have a capacity problem, and that is a different purchase.
Work out which one you have before you buy anything. Most phone systems will export a call log. Missed calls with a timestamp, one month, sorted by hour, answers the question in about ten minutes.
What an AI answering service actually handles well
Sorted by how reliable it is in production, best first.
- Capturing a caller's details accurately. Name, number, address, what they need, when. This is the thing AI does better than a rushed human, because it never gets bored on the fortieth call and never writes a number down wrong.
- Answering questions it has been given the answers to. Opening hours, service area, whether you handle a particular job, roughly what something costs. Grounded in your actual information, this is close to solved.
- Booking into a calendar. Reliable when the rules are simple. Gets fragile when availability depends on job type, travel time, and which van is free, which in a real service business it usually does.
- Qualifying and routing. Working out that this is an emergency, or out of area, or a supplier rather than a customer, and doing something different with it.
- Following up. The underrated one. A text within sixty seconds of a missed call recovers a meaningful share of callers, and no human answering service does this by default.
Where it fails, stated honestly
Four failure modes, and only one of them is really about the technology.
The caller is upset
Flood, break-in, medical, a job that has gone wrong. The AI will handle the conversation competently and the caller will still feel unheard. Route these to a human immediately, and build the detection for it deliberately rather than hoping the model works it out.
The answer is not in the system
A model with no grounding will produce something plausible. This is the failure that costs money, because a plausible wrong answer does not look like an error to anyone. It looks like an answer. Whatever you buy, ask what happens when the system does not know, and be unsatisfied with a vague reply.
Nobody maintains it
The most common real-world failure and it is entirely organisational. Prices change, the service area changes, a technician leaves. Six months later the AI is confidently quoting last year's price. An answering system is not something you install; it is something you keep current.
You did not have a phone problem
If your enquiries arrive mostly by form, message or referral, an answering service fixes a problem you did not have. We have talked people out of this more than once.
What it costs, against the human alternative
| Service | Type | Entry price | Billed by | Overage |
|---|---|---|---|---|
| Rosie | AI | $49/mo | Flat, 250 min | Tiered plans |
| Smith.ai (AI plan) | AI | $95/mo | Per call, ~60 calls | ~$2.40/call |
| PATLive | Human | $75–$235/mo | Per minute | $1.85–$2.60/min |
| Ruby | Human | $235/mo (50 min) | Per minute | $3.30–$5.40/min |
| Smith.ai (hybrid) | Human + AI | $292.50/mo (30 calls) | Per call | ~$9.75/call |
| AnswerConnect | Human | ~$325/mo | Per minute | $1.85–$2.95/min |
Published rates as of August 2026, taken from each provider's own pricing page or their published plan documents. Several of these vendors quote rather than publish, and every one of them changes pricing without notice. Treat this as a starting point for your own quote, not a guarantee.
The thing that decides your bill is not the headline price
Three billing models exist in this market and they behave completely differently once real calls start arriving. This matters more than the number on the pricing page, and it is the single most common reason a bill comes in at double what someone expected.
Per minute
Ruby, PATLive and AnswerConnect all bill this way. A five minute call costs five times a one minute call. The consequence is that your bill depends on how talkative your customers are, which is not something you control. A chatty caller describing a kitchen renovation costs you real money; a wrong number costs you the minimum increment.
Watch the overage rate specifically. Ruby's published overage runs from about $3.30 a minute on its largest plans to $5.40 on the smallest. That is the rate you pay for everything above your bundle, and the small plans have both the least included time and the worst marginal rate.
Per call
Smith.ai bills this way. A call is a call regardless of length, which makes the bill predictable and rewards you for letting conversations run properly. The trade is that short calls cost the same as long ones, so a business receiving lots of quick enquiries pays a premium.
Flat rate
Most of the AI-only services, Rosie among them, bundle a large block of minutes for a fixed monthly figure. This is the most predictable of the three and it is only possible because the marginal cost of an AI-handled minute is close to zero.
What did the median customer on this plan actually pay last month, including overage? Not the plan price. The real number. A provider confident in its pricing will answer.
How to decide, in about ten minutes
Export one month of call logs from your phone system. Filter to missed calls. Sort by hour of day.
- Clustered outside hours → a coverage problem. Almost any after-hours answering service fixes it, and the AI ones are far cheaper per call because nobody is being paid to sit awake.
- Clustered during business hours → a capacity problem. You are busy when customers call. AI helps, but so would a better routing rule, and you should price both.
- Spread evenly and low → you do not have this problem. Spend the money somewhere else.
The reason we push this exercise rather than a demo is that it produces a number you own. Every vendor in this market, us included, has an incentive to tell you the problem is larger than it is.
For most small service businesses the answering service is worker number one, not the whole system. The calls it captures are only valuable if they land somewhere that remembers them. An answering service that texts you a name and number has moved the problem, not solved it.
Answering service, or something that does more
Worth being clear about the distinction, because the pricing looks similar and the outcome does not.
An answering service catches calls. That is a bounded, well-understood job and the market for it is competitive, which is good for you as a buyer. If a caught call is all you need, buy one of the products in the table above and do not overthink it.
What we build is different in kind rather than degree: a system that holds the state of your business, so that the call, the follow-up, the quote and the job are one thread rather than four tools. That is a larger commitment and it is not the right purchase for everyone. On a platform we run in production, the measured cost of an AI-drafted reply is $0.00061, which tells you the model is never the expensive part. The expensive part is making it correct about your business and keeping it that way.
If you want the reasoning on that comparison in full, we wrote it up separately: what an AI employee costs compared with hiring one.
Common questions
What is an AI answering service?
An AI answering service picks up calls to your business using a voice model instead of a human operator. It greets the caller, works out what they want, captures the details, and either books the appointment, transfers the call, or sends you the message. Unlike voicemail it holds a conversation; unlike a human answering service it runs at a fixed cost and never queues.
How much does an AI answering service cost?
AI-only services generally start between $49 and $95 a month for a few hundred minutes or around sixty calls. Human answering services start meaningfully higher, typically $235 to $325 a month, and bill per minute with overage rates from roughly $1.85 to $5.40 a minute. Prices verified August 2026.
Is an AI answering service better than a human one?
For gathering structured information reliably at any hour, yes. For a distressed or unusual caller, no. The practical answer for most small businesses is neither on its own: AI handles the routine majority and anything it cannot classify goes to a person, which is why hybrid offerings exist.
Can callers tell they are talking to an AI?
Usually within a sentence or two, yes. The quality gap has narrowed but it has not closed. What matters more is whether the system says so. Being straightforward that it is an assistant, and that a person will follow up, consistently produces better outcomes than trying to pass as human and failing halfway through.
Will it work with my existing phone number?
Yes. The normal setup is conditional call forwarding: your existing number rings as it always has, and unanswered calls forward to the service after a set number of rings. You keep your number and nothing changes for callers who reach you directly.
Want to know what this would look like in your business?
No demo, no deck. A short call about which calls you are actually losing and whether answering them is worth automating. If it is not, we will say so.
Related: AI receptionist for small business · What answering services actually cost · What happens when AI gets it wrong
Want this running on your phones?
Tell us how many calls you miss in a week and what an average job is worth. If an answering service will not pay for itself on those numbers, we will say so.