HighLevel Voice AI Pricing: How the Per-Minute Billing Actually Works

HighLevel Voice AI is billed per minute of active call time, on top of your subscription. Understanding the structure matters more than memorising a rate, because the rate is adjusted more often than plan pricing and any number quoted on a blog goes stale.

How the billing works

Three facts define the model:

1. You pay for talk time, not for calls. A caller who hangs up after five seconds costs almost nothing. A four-minute conversation costs four minutes.
2. Billing rounds up in short increments, not to the full minute, so a 40-second call is charged as roughly 40 seconds rather than 60.
3. The charge has two layers. A voice engine component and language model token usage. The first is broadly fixed per minute; the second varies with how much conversation actually occurs.

Why the effective rate varies

Because token usage is one of the two components, the effective per-minute cost is not identical across calls. A conversation with long pauses and short answers consumes fewer tokens than one where the agent explains something at length.

In practice this means published rates and observed averages differ, and the average across an account will sit slightly above the headline voice engine rate. Budget against your own observed figure after a month rather than against a rate card.

Modelling it for your own volume

A method that gives a usable number in five minutes:

1. Count answered inbound calls per month. Your phone system already knows this. Use answered calls, not total, since unanswered ones are what Voice AI is replacing.
2. Find average call duration for the call type you plan to route to the agent. Booking calls are short; enquiry calls are longer.
3. Add a margin for AI conversations. They tend to run slightly longer than the human equivalent, because the agent confirms things a person would assume.
4. Multiply by the current rate shown in your own account.
5. Compare against the bundle. If the metered figure approaches the flat monthly fee, the bundle is the better structure.

Do this per sub-account rather than once for the agency, because the answer genuinely differs between a busy trades client and a quiet consultancy.

Bundle versus metered

The trap worth naming: if a meaningful share of your usage is outbound, the bundle covers less than it appears to, because outbound remains metered on top of the flat fee.

Controlling the cost

Since you pay for talk time, everything that shortens calls reduces the bill.

• Trim the qualifying questions. Three good ones beat seven mediocre ones, and each question costs seconds on every call.
• Give the agent a clear ending. An agent that does not know when it is finished keeps talking. Define the completion condition explicitly.
• Route obvious cases straight through. If someone says they want to book, book them. Do not qualify a caller who has already decided.
• Escalate early on complexity. A long conversation that ends in a transfer costs money and delivered nothing. Transfer sooner.
• Deploy after hours first. Bounded volume, and the comparison is with voicemail.
• Read transcripts for length, not just quality. The longest calls usually reveal a script problem you can fix once.

Setup guidance is in our Voice AI guide, and the bundle is covered in AI Employee.