What should I check in an AI voice agent's billing terms?
Seven things move the bill more than the headline price: how calls are rounded, whether spam and very short calls are billed, what a transferred call costs, whether failed calls count, what happens when you pass the allowance, how many calls can run together, and whether you pay for usage or outcomes. Two vendors with the same price per minute can produce bills 70% apart on the same calls. This guide sets out what each vendor publishes, read on November 12, 2026. Botcadence's own terms for its AI voice agent are included, the unflattering ones too.
Key takeaways
- Rounding: per-second billing is cheapest for short calls; rounding each call up to a full minute is the most expensive.
- Spam and short calls: Upfirst does not bill calls under 15 seconds, Goodcall counts unique callers, and Botcadence bills every answered call a 15-second minimum.
- Transfers: Bland AI charges 4 to 5 cents for each transferred minute; Retell AI and Botcadence stop the agent charge when the agent leaves.
- Past the allowance: some plans charge overage, Rosie moves you to the next plan, and Botcadence stops answering new calls until you add minutes.
- Concurrency and daily caps are limits on service, not only on cost. Ask for the numbers in writing.
1. Rounding and minimums
| Vendor | Published billing increment |
|---|
| Retell AI | Per second |
| PolyAI | Minutes are not rounded up, per its pricing page |
| Specialty Answering Service (live) | By the second |
| Botcadence | A 15-second minimum, then 30-second steps |
| Rosie | 60-second increments, rounded up, per its terms |
| AnswerConnect (live) | Rounded up to the next minute |
| Upfirst, Quo Sona, Smith.ai | Per call, whatever the length |
Take a line with 300 calls a month averaging 70 seconds. Billed by the second that is 350 minutes. With a 15-second minimum and 30-second steps each call bills 75 seconds, which is 375 minutes. Rounded up to the minute each call bills two minutes, which is 600. Same calls, a 71% spread.
2. Spam, silence and wrong numbers
Business lines attract robocalls, and how they are billed varies.
| Vendor | What its pages say |
|---|
| Upfirst | Spam, silent callers and calls under 15 seconds are not billed |
| Goodcall | Bills by unique caller, and does not meter minutes |
| Smith.ai | No charge for spam on its receptionist plans |
| Retell AI | Failed connections are not billed |
| Botcadence | Every answered call is billed at least 15 seconds; there is no spam exemption |
| ServiceTitan AI Voice Agent | Every call is billed whatever its outcome, per its help page |
If a fifth of your calls are junk, a per-call plan without a spam exemption is paying for them at full price. On a per-minute plan with a short minimum they cost little.
3. Transfers and hold time
When the agent hands a call to a person, three things can be billed: the agent's time while it dials and briefs, a transfer rate while your person talks, and the carrier's minutes for both legs.
| Vendor | Transfer terms |
|---|
| Bland AI | $0.05 per transferred minute on Start, $0.04 on Build, on top of the per-minute rate |
| Retell AI | The agent fee stops after the transfer; telephony continues |
| Botcadence | The agent's minutes stop once a person has taken the call and the agent leaves; time spent ringing your team is billed |
| Smith.ai receptionists | Extra transfer destinations are $15 a month each beyond those included |
With your own carrier, the forwarded leg is charged by the carrier at its normal rate.
4. Calls that never connect
Outbound campaigns place many calls that reach voicemail, a busy tone or nobody. Ask:
- Is an unanswered attempt billed?
- Is voicemail billed, and for how long does the agent stay on the line? Botcadence has optional voicemail detection that hangs up when it hears an answering machine; without it the agent leaves one short message.
- Is voicemail detection itself charged? Twilio lists it at $0.0075 a call and Telnyx at $0.002 to $0.0065.
- Does a batch-calling fee apply? Retell AI lists $0.005 a dial.
5. What happens when you pass the allowance
| Vendor | Past the included amount |
|---|
| Dialzara, My AI Front Desk, PolyAI | Per-minute overage at a published rate |
| Upfirst, Quo Sona, Smith.ai | Per-call overage at a published rate |
| Goodcall | Per-caller overage at a published rate |
| Rosie | Its terms say the account is moved to a higher plan |
| ElevenLabs Agents | Extra minutes at 8 cents; burst calls at 16 cents |
| Botcadence | New calls are refused until a 200-minute pack ($35) is added. A call in progress is topped up a minute at a time and ends with a closing line if no minutes are left |
| Ruby, Sameday | No overage rate on the pricing page |
The Botcadence term deserves attention: if you run out mid-month and nobody adds a pack, callers are not answered. Set a low-balance reminder, or buy packs before a busy period.
6. Concurrency and caps
A plan can have minutes left and still fail to answer, if too many calls arrive together or a daily cap is reached.
| Vendor | Published limits |
|---|
| Bland AI | 10 concurrent lines and 100 calls a day on Start; 50 lines and 2,000 a day on Build |
| Retell AI | 20 lines included, then $8 a line a month |
| Vapi | 4, 10 or 30 lines by plan, then $10 a line a month |
| ElevenLabs Agents | 4 to 40 lines by tier, with paid burst above |
| PolyAI | 5 or 10 lines by plan |
| Botcadence | Set per account. It starts at one line; inbound calls over the limit are rejected before they are answered, and campaign calls wait in a queue |
One line is enough for a quiet office and far too few for a campaign or a storm day. Ask what your limit is, how it is raised and how long that takes.
7. Usage pricing or outcome pricing
Most voice agents charge for usage: minutes, calls or callers. A few products charge for outcomes, such as a resolved conversation or a booked appointment. Dialpad's pricing page describes AI agent credits charged per conversation and says you are not charged if no work is done.
| Usage pricing | Outcome pricing |
|---|
| You pay for | Time or calls | Results the vendor defines |
| Predictable when | Call volume is steady | Outcomes are steady |
| Watch for | Rounding, spam, overage | The definition of an outcome, and who decides whether one happened |
Neither is better in general. Outcome pricing needs a definition you can audit from your own records.
Questions to put to any vendor, in writing
- What is the billing increment, and the minimum for a short call?
- Are spam, silent and failed calls billed?
- What does a transferred call cost, on each leg?
- What happens when I pass my allowance, and will calls stop?
- How many calls can run together, and what does raising that cost?
- Are there daily or hourly caps?
- Are the language model, speech, carrier and numbers inside the price?
- What did a customer like me pay in their busiest month?
For the per-minute arithmetic see AI voice agent cost per minute, and for plan prices see the AI voice agent pricing guide. The AI receptionist for small business page shows what Botcadence's plan includes.