Voice agent
The agent that answered: speech in, a decision, speech back out, plus any tool it had to call to answer the question. This is the base of every metered minute.
Usage-based pricing
Zumu’s AI call center pricing is usage based and nothing else. You pay per minute for the minutes the AI agent actually handled, metered in five components you can see on every call, against a rate card that belongs to your organization and carries the date it took effect.
These are the platform’s list rates, the same card every new organization starts on. Volume and committed-use pricing is a conversation, not a discount schedule. Some organizations run on a rate card negotiated for their own call mix. Ask us whether your own rate differs from this one.
One call, metered
A caller phoned in, the agent handled the first eight and a half minutes, and then a specialist on your team took the call. Five components meter against the same seconds, and all five stop on the same line. What happens on the other side of that line is recorded, transcribed and reported, and none of it is billed as AI minutes.
Session receipt
call 4f21 · scheduling
Call duration
One call, in the shape the metering service reports it
08:32
08:32
08:32
08:32
08:32
Warm transfer at 08:32. Your specialist takes the call.
Human leg 05:30, not billed as AI minutes05:30
Metered
08:32 of 14:02
Multiplied by
your organization’s dated rate card
Reported as
a cost line on this call, to the cent
How metering works
Every metered minute is the same minute counted by five services that each did something you can name. There is no bundle to take apart afterwards, and no line on the card that exists because it was easier to price that way.
The agent that answered: speech in, a decision, speech back out, plus any tool it had to call to answer the question. This is the base of every metered minute.
Background noise removed from the caller’s side, so somebody calling from a depot forecourt is still understood the first time they say it.
Per turn timings, silent turn and interruption counts, struggle signals, and the session report that lets you reconstruct the call weeks later.
The audio itself, kept with per participant tracks so you can replay the caller and the agent separately rather than as one mixed file.
The phone leg. It is metered only on calls that arrive over a phone number, so a call taken through the widget in a browser does not carry it at all.
Current list rates
Phone calls
$0.156/min
All five components meter, including the phone leg.
Browser or widget calls
$0.146/min
Four components meter. There is no phone leg to charge.
What you see on this page is what appears on the invoice. Every call’s session report itemizes the same five components, to the cent.
Rate cards are scoped to your organization and effective dated. A change is a new card with a date on it rather than a quiet edit to the one you are already on, and old invoices stay reproducible against whichever card was in force when the call happened. Your rate never moves behind you.
The session report for a call itemizes cost by component beside the transcript, the tool calls and their durations, which provider endpoint actually served the call, and how long the caller waited before the agent spoke. A charge you cannot explain is a charge you can open.
One clock, five components, and a rule that stops the meter when a person takes over. That is the whole billing model.
Try it yourself
Move the sliders to your own call volume, average call length and phone-versus-widget mix. The math runs on the same five rates published above.
60% phone · 40% browser or widget
Volume and committed-use pricing is a conversation. Talk to us
Estimate at list rates · your rate may differ
Estimated monthly total
$760
$0.152/min effective, across 5,000 min
The transfer rule
Metering bills the seconds the AI agent actually handled and stops there. The minutes after the handoff are recorded, transcribed and reported, and none of them reach the invoice. The rule lives inside the metering service rather than in somebody’s memory at the end of the month, which is why it holds on every call rather than on the calls anyone thought to check.
Measured, one operator
~36%
of one operator’s weekly bill was correctly excluded, because their own team handled those minutes.
Treat that figure as proof the rule works rather than as a forecast of your invoice. Your share depends entirely on how often your calls end up with a person, and your call mix will not be theirs.
Where the figure comes from
The handoff itself is the part most platforms give up on. See how a warm transfer runs
What is included
No feature gates and no tiers to compare. Every organization runs on the same platform, which is why the only quantity on your invoice is minutes.
There is no plan that keeps the wallboard behind an upgrade. The only thing that moves your bill is how many minutes the AI handled.
Questions
Bring last month’s call volume. We will show you the rate card and where the meter would have stopped on your own transfers.

Talk to us
Bring a week of your call volume and the mix behind it. We will walk the rate card line by line, show you where the meter stops, and let you hear the agent take a call before anything is agreed.