How human services actually bill, and where the surprises are
The advertised model is a monthly plan with an included allowance of minutes or calls, plus a per-minute rate beyond it. At the time of writing, small business plans commonly advertise somewhere in the low hundreds per month, with per-minute rates on top, but these figures move and vary by market, so treat any number you read, including ours, as a starting point for a quote rather than a fact.
The line items that surprise people are not in the headline price: setup fees, overage rates once the allowance is gone, the rounding increment (billing in one-minute blocks costs meaningfully more than billing by the second on short calls), holiday and after-hours surcharges, and sometimes a charge for calls that were not for you at all.
The structural consequence matters more than any individual fee. Your bill rises with your call volume, which means it rises fastest in exactly the weeks when you are busiest, and a heat wave or a cold snap arrives as a bill as well as a workload.
The short versionAsk for the rounding increment, the overage rate and the surcharge list. The headline plan price is not the price.
How AI agents bill, and where their surprises are
Usually a flat monthly fee, sometimes with a usage component, and for a custom build a one-off cost to create the thing. The cost does not climb with call volume in the same way, which is the genuine structural advantage: the busy week is free.
The surprises here are different in kind. Integration is where budgets actually go, because an agent that cannot see your live availability and write into your schedule is a very expensive voicemail. Ask what happens when the calendar integration fails. Ask whether the agent can be updated when your prices or service area change, and by whom.
The other cost is your time up front. A good build needs someone from your business to explain how the calls actually work, and that is unavoidable rather than a sign of a bad vendor.
The short versionFlat cost is the real advantage. Integration is where the money and the risk actually sit.
The variable that decides it, which nobody states
What share of your calls need genuine human judgement?
Not "would be nicer with a human", but genuinely cannot be handled by a system following rules: an unusual commercial diagnostic, a distressed customer, a negotiation, a situation nobody wrote a rule for. Estimate that percentage honestly for your own call mix, because it decides the answer more than price does.
If it is high, a human service is worth its cost and the AI option will frustrate everyone. If it is low, which it is for most residential service businesses where the same fifteen situations repeat, you are paying a premium for judgement you rarely use.
Almost nobody in this market asks the buyer that question, because for a vendor it is a question with a wrong answer.
The short versionEstimate the share of calls needing real judgement. High means human, low means AI. That is the decision.
What humans are genuinely better at
Anything unscripted. A caller who is upset, a situation that does not fit a category, a conversation that needs judgement about what the person actually wants rather than what they said.
They also handle the caller with no vocabulary better. Someone describing a noise rather than naming a component is easy for an experienced person and hard for a system, though this is the gap that has narrowed most.
And there is a real, hard-to-quantify value in a customer knowing they spoke to a person, particularly for higher-value work and older customer bases.
The short versionUnscripted, distressed and ambiguous calls are the human case, and it is a real one.
What AI agents are genuinely better at
Never being unavailable. No rota, no holiday, no sick day, no queue at three in the morning during the first cold snap, which is precisely when a human service is most likely to have you holding.
Consistency. The hundredth call of the night gets the same diagnostic questions as the first, which is not true of tired people, and it is the reason the captured detail is usually better.
Knowing your business specifically. A human answering service operator is handling several companies and works from a script about yours. An agent built for you knows your equipment mix, your service area and your actual availability, and books into it directly rather than promising a callback.
The short versionAvailability, consistency and depth of knowledge about your specific business.
The hybrid, which is what most people should actually buy
The framing as a binary choice is mostly a vendor artefact. The arrangement that works is an agent handling first contact on every call, resolving the routine majority, and escalating the exceptions to a person, whether that is your own team or a human service kept for the overflow.
This also fixes the cost curve. You are paying humans only for the calls that needed humans, rather than for every call in order to catch the few that did.
If a vendor tells you their product should handle one hundred percent of your calls, that is a claim about their sales target rather than your business.
The short versionAgent first, human for exceptions. You stop paying for judgement on calls that never needed it.
When we would tell you not to buy from us
If your call volume is low enough that you or someone in the family genuinely answers every call, there is nothing to fix and both options are worse than what you have.
If most of your work is complex commercial diagnostics where every call is different, the judgement share is high and a human service is the better purchase.
And if your scheduling lives entirely in someone's head or on paper, the integration that makes an agent valuable does not exist yet. That is a scheduling project first and an answering project second, in that order.
The short versionLow volume, high judgement share, or no schedule to integrate with. In those three cases, do not buy an agent.
