AI vs. human phone banks: cost per contact

A defensible cost-per-contact comparison for campaigns weighing AI voice against human phone banks — what each layer actually costs, and where the meter hides markup.

Every campaign that has ever run a voter-contact program knows the phone is where persuasion and turnout still happen — and knows how expensive it gets. A human phone bank has a hard floor set by wages; an AI voice program has a much lower floor set by a commodity technology stack. The honest comparison is not “robot versus person” but cost per completed contact, measured the same way on both sides, with the markup pulled out into the open.

What a human phone bank actually costs per contact

A paid phone-banker reaches a modest number of live voters per hour after bad numbers, no-answers, and refusals. Blend the fully loaded wage, the dialer, list costs, and supervision, and the cost per meaningful conversation typically lands in the dollars, not cents — before a single voter is persuaded. Volunteer banks trade cash for coordination overhead and inconsistent scripting. Neither scales on the last weekend, which is exactly when you need surge capacity.

The other cost is variance. Human quality drifts across a shift, scripts get improvised, and you often cannot prove afterward what was said on any given call. For a regulated activity, that is a compliance liability as much as a quality one.

What an AI voice call actually costs to run

A production voice call runs on speech-to-text, a language model, text-to-speech, telephony, and hosting. Added up, that stack costs roughly 4–5¢ per minute; a five-minute call is about $0.21–$0.26, and a web-based call is closer to 3¢ per minute. That is the real floor. Anything you pay above it is platform value or platform markup — and it is worth knowing which.

• Speech-to-text (transcription of the caller)
• The language model that decides what to say
• Text-to-speech (the voice the voter hears)
• Telephony minutes and carrier fees
• Hosting and orchestration

Because that floor is a commodity, a vendor charging a fat per-minute rate is often selling you a small markup on the same components everyone else uses. Per-minute-only pricing quietly surrenders the value of the outcome to a meter. A hybrid model — a platform fee plus usage — keeps the meter honest and prices the thing you actually bought: a completed, logged, compliant contact.

The comparison campaigns should actually run

Do not compare raw minute rates. Compare cost per completed contact, and hold both sides to the same compliance bar: AI self-disclosure at the top of the call, all-party recording consent where required, and a contact-frequency ledger so you never over-call the same household across channels. An AI program clears the last-weekend surge that human banks cannot, and it can prove what it said on every call — a record a human bank rarely produces.

Price the completed, disclosed, logged contact — not the minute. AI’s cost floor is a commodity 4–5¢/min; the value is a program that surges on the final weekend and can prove what every call said.