How to charge clients for AI voice minutes

Voice minutes are a variable cost, and unlimited plans turn them into your risk. Here is how to bundle an allowance, set an overage rate, and check real usage at 60 days, with the math on a $0.35 wholesale minute.

Lumina8 min read
Abstract amber arcs and a fine waveform ending in evenly spaced tick marks on a near-black charcoal background

Bundle a minute allowance into the client's monthly price, set a per-minute overage rate at roughly double your own cost, and look at real usage after 60 days. Do not sell unlimited minutes. On Lumina Sales, voice minutes on a client build cost you $0.35 each wholesale, so an "unlimited" promise is an open-ended bill that you signed and the client did not.

That is the short answer to a question every agency hits sooner or later: how should I charge a client for AI voice minutes? The rest of this post shows the arithmetic, the three ways people do it, and the client for whom none of this works.

What does a voice minute cost you?

On a Lumina Sales client build, voice is billed per minute on top of the setup and monthly. The published wholesale guide lists Voice AI on its own at $450 setup and $45 a month plus $0.35 a minute. The Top Tier package, which bundles the website, chatbot, follow-up, SEO and voice, runs $315 to $460 a month plus the same $0.35 a minute. Final numbers get agreed on the 30-minute pricing call before anything is built.

Those are guides to what you pay us. Your client never sees them.

It helps to compare that to the raw platform cost, because a sharp client or a developer friend will. Retell lists voice infrastructure at $0.055 a minute, text-to-speech from $0.015, telephony at $0.015, and a language model on top that ranges from a fraction of a cent to several cents a minute. Its own sample configuration comes to $0.11 a minute. Vapi charges $0.05 a minute for hosting and passes model, voice and transcription through at cost, and its worked example puts 1,000 minutes at $82 to $129. We covered the full breakdown in what a voice AI agent costs per minute.

So $0.35 is above the raw usage cost. We would rather you hear that from us. What it buys is no API keys, no phone numbers to provision, no concurrency to manage, and a build that someone else maintains. If you want to run voice infrastructure yourself, the minutes are cheaper and the hours are yours.

How many minutes will a client use?

Guess before you quote, then check against reality at 60 days. Here is a working estimate for a small service business, and the assumptions are ours, not a measured average.

A plumbing or HVAC shop that gets 15 calls a day the AI actually handles, at about a minute and a half each, over 22 working days, is roughly 500 minutes a month. At $0.35 that is $175 in minutes. A dental practice with a lighter after-hours load might sit near 200 minutes, or $70. A busy personal injury intake line with long, emotional calls could run past 1,000 minutes, or $350 and up.

Those ranges are wide, which is the whole problem. You are pricing something whose volume you do not control and cannot see until it runs. One vendor in this space, Dograh, offers a rule of thumb that is useful even though it sells a competing product: usage-based billing suits volume that swings more than 30 to 40 percent month to month, and a flat fee suits volume that holds within a band of 10 to 15 percent. Plenty of local service businesses swing with weather, season and ad spend.

What are the three ways to charge for minutes?

Pass-through with a markup

You bill minutes as a separate line: $0.70 a minute, itemized on the client's portal. Your margin on minutes is a clean 50 percent at any volume, and you carry no risk.

The cost is sales friction. The client now has a meter running, and owners who are already nervous about AI will ask what happens if it goes viral. Pass-through also trains them to watch minutes instead of booked jobs, which is the wrong number to stare at.

A bundle with an allowance and an overage rate

You include a set number of minutes in the monthly price and charge a per-minute rate beyond it. This is what we recommend for most clients, and buyers already understand it. Smith.ai's published plans for its human receptionists work this way: the Starter plan is $300 a month for 30 calls, with $11.50 for each call over.

Here is how it plays out on a Top Tier client. Say your wholesale monthly is $400, inside the published range, and you sell the buildout at $1,500 a month with 300 minutes included.

Client usesYour minute costYour total costOverage billed at $0.75You keep
300 minutes$105$505$0$995
600 minutes$210$610$225$1,115
1,000 minutes$350$750$525$1,275

Notice what happens as usage climbs. You keep more, not less, because the overage rate sits above your cost. The client who talks to the AI the most is also the client getting the most value, and they are paying for it in proportion. Your platform subscription of $49 is separate and does not scale with any of this.

The allowance needs to sit where a typical month lands under it. If most months run over, the client feels nickel-and-dimed. If it is far too generous, you gave away margin for nothing. Set it near the 70th percentile of your estimate, not the average.

Unlimited

Do not do it. The wholesale minute is a variable cost, and unlimited turns it into a fixed price against an unbounded bill. One client with a lead-gen campaign pointed at the phone number, or a voice sales agent running outbound dials, can use five times what you modeled. You will find out when the invoice arrives, and the client will have no reason to slow down.

If a prospect insists on unlimited, quote a very high allowance with a fair overage instead. The word "unlimited" costs you nothing to drop and a lot to keep.

Where this approach breaks

Minute pricing is not the right tool in a few cases, and you should know them before you build a quote around it.

High-volume lines are the first. At several thousand minutes a month, the gap between $0.35 and the raw platform cost gets large in absolute terms. At 5,000 minutes, wholesale minutes alone are $1,750, while a self-assembled stack at Retell's sample $0.11 would be around $550 before your own labor. If you land a client like that, price overage tightly, or decide whether that one account deserves its own infrastructure.

Outbound is the second. Inbound call volume follows the client's customers. Outbound volume follows a campaign someone configures, which means a mistake can burn minutes quickly. Price a voice sales agent with a firm monthly cap and ask for approval above it.

The third is the demo allowance. The 100 voice minutes in your $49 plan exist for selling. They are not production capacity for a client, and the client builds are billed separately. We wrote up how far the demo allowance goes in will 100 voice minutes cover your demo pipeline.

Finally, this is for you if you want a recurring margin on voice that scales with the client's success. It is not for you if you need a single flat number to close a prospect who will walk at any variable cost. In that case, sell the chatbot and website tier first and add voice at the 60-day review.

How should you present minutes to the client?

Show it as a line in the proposal: "Includes 300 call minutes a month. Additional minutes are $0.75." One sentence, no table, no vendor names. Then tell them where they can see their usage. The owner who can open a dashboard and see a count will rarely dispute it.

Put a date on the calendar to revisit at 60 days. Pull the real minute count, compare it to your estimate, and either raise the allowance or adjust the monthly. Doing this once, early, is what keeps a pricing mistake from turning into a year of lost margin. The structure of that first conversation is in our post on the 30-minute pricing call for client AI builds.

FAQ

Should I show my wholesale cost to the client?

No. Your client buys an outcome at a price you set, and the wholesale number is your cost of delivery. Itemize minutes on the client side at your retail rate. Nothing about the Lumina Sales structure requires disclosing what you pay.

What overage rate should I charge?

Start around double your cost, which is $0.70 to $0.75 a minute on a $0.35 wholesale minute. Go lower if your allowance is tight and the client is price sensitive, and higher if they are likely to run well past the bundle. The rate matters less than having one written down.

What happens if a client blows through their minutes?

The calls keep being answered, and you bill the overage at the end of the month. Tell the client early when they pass 80 percent of the allowance so the invoice is never a surprise. If usage is consistently double the plan, move them to a larger bundle with a higher monthly.

Do the minutes in my $49 plan cover client calls?

No. Those 100 minutes cover your own demos. Voice on a client build is billed per minute on the build, agreed on the pricing call before anything is built.

Can I just raise the monthly and skip the overage?

You can, and for a stable, low-volume client it is simpler. The risk is that you set the monthly against a guess and absorb the difference when reality runs high. The overage line is cheap insurance.

Next step

Open your most recent client proposal and add one line: an included-minutes number and an overage rate. Then take the live numbers to your next pricing call, where the setup, monthly and per-minute rate get fixed before the build starts. You can start your agency workspace for $49 a month and try the live demo first to hear what your clients' callers would hear.