Every voice platform sells the handoff. The agent takes the call, handles what it can, and passes a human the ones it cannot. It is the feature that makes the category safe to buy.
Almost none of them tells you what the handoff costs.
Bland bills the transfer twice, and publishes the example
Bland is the exception and deserves credit for it. Its documentation carries a worked example: a ten-minute call where the agent talks for eight minutes and then transfers to a human for two.
On the Scale tier that bills ten minutes of connected time at $0.11 plus two minutes of transfer time at $0.03, totalling $1.16. The two transferred minutes are counted twice, once inside the connected total and once again as transfer.
Divide it out and the effective rate on that call is $0.116 a minute against a $0.11 headline. On a ten-minute call with a two-minute handoff, that is about a 5% premium, and it scales with how often your agent gives up.
There is a second finding buried in the same table. Customers who bring their own Twilio account pay no transfer fee at all. That turns the carrier decision from a plumbing question into a rate question, and it is not framed that way anywhere on the pricing page.
Ultravox: legs that outlive the agent
Ultravox publishes a different and equally consequential warning. Its SIP minutes can accrue while your agent is not connected to the user, on transfer legs that outlive the agent.
Picture what that means in practice. The agent transfers, drops off, and the caller sits with a human for twenty minutes. If the leg is still open on your account, you are paying carriage for a conversation your software is no longer part of.
Regal: the mechanics documented, the billing silent
Regal builds the best warm transfer in this directory. The agent holds the contact, waits for the human, passes a summary, then drops, with recording continuous throughout. The documentation covers cold and warm mechanics in real detail.
It says nothing whatsoever about billing. On a product built around handoff, that is the single most consequential silence on the site, and it is the question to put in writing before signing.
The staffing arithmetic underneath all of this
Regal also publishes the calculation nobody else does, and it is about people rather than minutes. Fifty concurrent AI calls, a five-minute average, a forty percent transfer rate and a thirty-minute human close require about 120 licensed human agents.
The transfer rate is the variable that hurts. At forty percent, the AI is not replacing your phone team; it is triaging for a phone team you still need to employ. Anybody modelling savings from an AI voice deployment should run that calculation against their own transfer rate before signing anything, because it is the number that decides whether the business case survives contact with a real call queue.
What to ask
- Is a transferred minute billed once or twice? Ask for the worked example, because one vendor publishes one.
- Does the transfer leg count against my concurrency limit, and does it keep counting after the agent drops?
- Does bringing my own carrier remove the transfer fee, as it does on at least one platform here?
- What is the expected transfer rate for my use case, and what headcount does that imply?
- If the transfer fails, does the caller get a queue, a voicemail, or a dial tone?