Pass-through voice pricing: the math that lets you quote a flat client rate.

Agency pricing for voice AI has converged on a simple, defensible structure: charge the client a flat rate per minute of agent time. Twenty-five cents, thirty cents, sometimes a buck for high-touch verticals. It's clean on the invoice, easy to forecast for the client, and matches how the client thinks about phone work generally.

The trouble is that what you charge and what you pay are two different lines on two different ledgers, and most agencies optimize one without doing the work on the other. If the cost ledger moves at call time and the price ledger doesn't, the spread between them is going to bend in a direction you don't want.

The two postures, in one paragraph each

Metered cost. Your infrastructure charges you per-token at call time. A heavier caller costs more. A longer call costs more. A bigger knowledge base costs more, every minute, for every call. You quoted the client a flat rate; you're now absorbing volatility you can't predict at the moment a robocaller dials in.

Flat / pass-through cost. Your infrastructure charges you a fixed per-minute rate that you locked at build time when you picked the agent configuration. A robocaller costs you the same as a qualified buyer. A 14-minute conversation costs you exactly fourteen times what a one-minute conversation costs. The caller has no leverage on your bill.

The first structure is fine if your volume is low, your callers are predictable, and your retention is short enough that one bad month doesn't reset your unit economics. The second structure is what you want when you're building an agency you intend to operate at scale for years.

The actual margin math

Let's run a real example. You sell a client 1,000 minutes a month at $0.30/min — $300/mo in voice revenue. Your infrastructure cost is one of the two postures.

Flat infrastructure

• Cost: $105/mo (1,000 × $0.105)
• Revenue: $300/mo
• Margin: $195/mo — locked at the time of build
• Variance month-to-month: ~0

Metered infrastructure (real measurements)

• Cost on a light agent month: $142/mo (1,000 × $0.142)
• Cost on a heavier agent month: $215/mo (1,000 × $0.215)
• Margin: $85 to $158 — moves with caller pattern
• Variance month-to-month: 86% swing between best month and worst

The flat-infrastructure version banks $195/mo per account, every month, predictably. The metered version banks somewhere between $85 and $158, and the agency doesn't get to choose which end of the range. That's the same client, same product, same price. The only thing that changed is which side of infrastructure the agency is sitting on.

What "pass-through" actually means

Pass-through is a specific posture, not a marketing word. It means the platform is charging you the wholesale rate it pays to the underlying providers (LLM, voice synthesis, telephony) without taking a markup, and surfacing the cost as a flat per-minute line so it doesn't move at call time.

On Callibre's Voice AI line, that decomposes as:

• $0.07/min Voice AI — STT, TTS, voice orchestration. Flat. Locked.
• $0.006–$0.060/min LLM — you pick the model at build time. Whatever you pick is the rate.
• $0.010/min telephony — pass-through at the carrier rate. Same vendors anyone uses.

Add the three and you get a number between $0.086 and $0.140/min, depending on the model. Pick once, the rate locks, and your only cost-side variance is whether you change the agent — not what the caller does on a given call.

Why agencies want this on the cost side specifically

Agencies sell outcomes. The client doesn't care which LLM is running underneath; they care whether the agent books appointments and answers questions properly. They want a flat price because predictability is part of the product.

If your cost is also flat, you can defend the flat price you quoted, with margin, for the full term of the contract. If your cost moves and your price doesn't, the gap goes one direction every time — toward you eating cost the client doesn't see.

The agencies that successfully scale past $50K MRR in voice AI almost all converge on flat infrastructure for this exact reason. Not because flat is cheaper on every minute (it sometimes isn't — a very light caller on a metered platform can come in below a flat rate). It's that flat is defendable, and defendable margin is what compounds over time.

The simpler version

If you take one thing from this post: match the shape of your cost to the shape of your price. Flat client price + metered infrastructure is the most common pricing mistake in voice-AI agency work. The fix is moving infrastructure to flat. That's the structural shift Callibre exists to make available.