Unlimited minutes don't exist: how many voice minutes your business actually needs
When comparing quotes for an AI voice agent, the number that matters isn't the monthly fee: it's the price per minute and how your usage is billed. A quote promising unlimited minutes with no number written beside it isn't a generous offer — it's a number someone will have to correct later.
Why voice is priced differently from WhatsApp
A WhatsApp message has a negligible marginal cost: text in, text out. A phone call does not. Every minute of conversation consumes speech transcription, response generation, speech synthesis and telephony — four pay-per-use services which today add up to roughly ten to fifteen cents per minute at provider cost.
That means the cost of the service grows with your call volume. Any flat voice rate either hides a limit in the small print, or turns the best customer (the one getting the most calls) into the worst deal for whoever sold it. Someone pays the difference in the end.
That's why Noema bills the voice channel on real minute usage, with a monthly minimum covering availability, maintenance and support. No plans, no bundles: a quieter month costs you less. It's less impressive than "unlimited" and it's what keeps the service alive a year from now.
Estimate your minutes in three steps
Step 1: incoming calls per day. Not answered calls — incoming ones. If you don't know, your landline or switchboard log does.
Step 2: average length of an admin call. Booking a slot, moving an appointment or asking a price are short conversations; two to three minutes is a prudent estimate.
Step 3: multiply by open days per month. Example: 20 calls a day × 3 minutes × 22 days gives 1,320 minutes a month. If the agent only picks up what reception can't reach, the real figure will be a fraction of that; if the agent takes every call, it will be that or more.
| Scenario | Calls/day the agent takes | Average length | Minutes/month |
|---|---|---|---|
| Out of hours and lunch only | 5 | 3 min | ~330 |
| In-hours overflow + out of hours | 10 | 3 min | ~660 |
| Mid volume, every call | 20 | 3 min | ~1,320 |
| High volume, every call | 40 | 3 min | ~2,640 |
Hybrid architecture: the agent doesn't have to take every call
The usual request is "have the agent take every call". It's understandable and it's almost always overpaying. The setup we recommend is different: during opening hours, conditional forwarding — the phone rings at reception and the agent only picks up what nobody answers. Outside hours, over lunch and at weekends, the agent takes everything.
The outcome the customer perceives is identical ("we don't miss calls") with substantially lower minute consumption. And the share of volume that does consume minutes is precisely the share that used to be lost.
It's the same logic we use to work out the real cost of a missed call: the point isn't covering hours, it's covering the hours where customers are lost.
Four things to check in any voice quote
1. How many minutes the plan includes, written as a number.
2. What a minute costs. If it says "at provider cost" or says nothing, that's an open-ended figure.
3. What counts as a minute: only calls the agent handles, or also calls it transfers to a human.
4. Whether what you pay tracks your real usage. It's the only honest way to close an estimate nobody can nail on day one: with usage-based billing, getting the forecast wrong hurts nobody.
Frequently asked questions
Why does nobody genuinely offer unlimited voice minutes?
Because every call minute has a real provider cost (transcription, model, speech synthesis and telephony). Under a flat rate, the customer receiving the most calls leaves the least margin: either there is a limit in the small print, or the service isn't sustainable.
How many minutes does a typical clinic need?
It depends on volume and on whether the agent takes every call or only the overflow. With the hybrid setup we recommend, a mid-volume clinic usually sits below a thousand minutes a month. The way to know is to look at one month of call logs.
What happens if one month brings far more calls than expected?
The invoice rises proportionally and predictably, at the same per-minute price as always: no steps, no penalties, no separate overage rate. What should never happen is the per-minute price being unwritten or pegged to provider cost: that's an open-ended figure.
What if real volume differs from the estimate?
Nothing happens, and that's precisely why we bill on usage. No estimate made on day one is exact; under a fixed fee, getting it right or wrong hurts one of the two parties. Paying for real usage takes the gamble out of the forecast.
Want the estimate with your own numbers?
Bring one month of call logs and we'll tell you how many minutes you actually need, before you sign anything.
Free demo