AI Voice Agent Cost: A Real Anatomy, No Hand-Waving

AI voice agent cost comes in two parts that people constantly mix up: what it costs to run per call, and what it costs to build. The running cost is cents per minute. The build is where the real number lives, and it ranges from a cheap monthly subscription to a five-figure custom project. Let’s take both apart honestly.

We build these for clients and we run our own — Mercvox, an AI receptionist for trades businesses across North America. So when we talk about cost layers, we’re talking about bills we actually pay.

The three running-cost layers

Every AI voice agent, no matter who builds it, runs on three stacked costs. Per minute of conversation, you’re paying for all three at once.

Layer one: the AI model. This is the brain — the language model that understands the caller and decides what to say. It’s metered by tokens, which roughly track how much gets said. For a normal back-and-forth voice call this tends to land somewhere around a few cents a minute, depending on which model and how much context you feed it.

Layer two: voice synthesis. This is turning text into a natural-sounding voice, and turning the caller’s speech back into text. Better, more human voices cost more. Realistically this is another couple cents a minute, and the gap between a cheap robotic voice and a genuinely good one shows up right here. If voice quality matters to you — and it should — this is not the layer to cheap out on.

Layer three: telephony. This is the actual phone line — carrying the call over the network, the phone number, the per-minute carrier charge. Providers like Twilio price this, and it’s usually around a cent or two a minute plus a small monthly number fee.

Stack the three and a typical voice call runs somewhere in the range of roughly a nickel to fifteen cents a minute in raw cost, before anyone’s margin. That’s the floor. Anyone quoting you wildly less is cutting one of these layers — usually the voice.

Off-the-shelf tools vs a custom build

There are two ways to get an AI voice agent, and they’re priced completely differently.

Off-the-shelf subscription tools are the SaaS route. You sign up, configure a few things in a dashboard, and pay a monthly fee — often a few hundred dollars a month for a small business tier, with usage on top. They’re fast to start and cheap up front. The catch is they’re generic. You get the agent the platform gives you, the integrations they support, and the limits they set.

Custom builds are what we do. Someone designs the call flows around your actual business, wires the agent into your specific software, handles your edge cases, and tunes the voice and guardrails. This is a project with a real build cost — anywhere from a few thousand dollars for something simple to well into five figures for something deeply integrated — plus the running per-minute costs above.

Neither is “better.” A solo operator with simple needs is often well served by an off-the-shelf tool. A clinic juggling insurance questions, a multi-location trades company, or anyone whose calls touch sensitive data usually needs the custom route. We dug into this tradeoff in custom AI agent vs off-the-shelf chatbot.

What drives a custom build up or down

When we scope a custom voice agent, a handful of things move the number more than anything else.

Integrations. An agent that just answers FAQs is cheap. An agent that reads your live calendar, books into Jobber or ServiceTitan, checks availability across locations, and writes back confirmations is doing real engineering work against real APIs. Every system it has to talk to adds cost — and adds most of the value.

Channels. Voice only is one build. Voice plus SMS plus web chat is three surfaces that all have to stay in sync. If you want the agent to text a confirmation after the call, that’s another integration — the kind of thing we cover in WhatsApp and SMS automation for small business.

Compliance. If your calls touch health data, payments, or anything regulated, the build needs guardrails, logging, and handling that a generic agent doesn’t. A dental clinic taking patient details has a higher floor than an auto shop booking oil changes — we get into the specifics in our piece on AI receptionists for dental clinics.

Call complexity. How many different things can a caller ask for, and how messy are the edge cases? More branches, more tuning, more cost.

Why cheap DIY builds get expensive later

This is the part nobody warns you about. You can stitch together a voice agent yourself from open tools over a weekend and feel like you saved a fortune. Then the bills start.

The cheap voice sounds robotic, so callers hang up and you’re paying for calls that don’t convert. The agent makes things up because nobody built guardrails, and now it’s confidently quoting wrong prices — see how to stop AI agents making things up for why that happens. The integration breaks every time your scheduling software updates, and there’s no one to fix it but you. The thing that took a weekend to build takes a day a week to babysit.

Cheap up front, expensive to own. The real cost of a voice agent isn’t the build — it’s the maintenance you didn’t price in. A good custom build front-loads that work so the thing actually runs without you.

How we quote at Driftel Labs

We don’t give cold estimates. Someone asks “how much for an AI voice agent” and the honest answer is always “it depends on what it has to do,” and a number pulled out of thin air helps nobody.

So we run a written discovery first. We map your call types, the software it needs to touch, your volume, your edge cases, and your compliance needs into a short plan you can read. Then we give you a fixed quote against that plan. No surprise change orders, no meter you can’t see. You know the number before we build anything, and the number is tied to a document that says exactly what you’re getting.

We think this is the only fair way to do it, because the thing that determines cost — what the agent actually has to do — is the thing a cold estimate ignores. You can see the kinds of businesses we build for on our industries page.

What you should expect to pay

Honest, wide ranges, because anyone giving you precise numbers without knowing your setup is guessing.

Running cost: roughly a nickel to fifteen cents a minute of conversation, stacking the model, voice, and telephony layers. A business doing a few thousand call-minutes a month is looking at usage in the low hundreds of dollars.

Off-the-shelf tools: commonly a few hundred dollars a month for a small business, usage on top, fast to start, generic in fit.

Custom build: a one-time project cost from a few thousand dollars for something simple to well into five figures for deep integrations and compliance, plus the running costs and ongoing support. The integrations and channels you need move this more than anything else.

For the bigger picture of what one of these does for a service business once it’s running, here’s our overview of AI voice agents for service businesses.

FAQ

What’s the cheapest way to get an AI voice agent?

An off-the-shelf subscription tool is cheapest up front — a few hundred dollars a month, running in a day. Just go in knowing it’s generic, the integrations are limited, and you may outgrow it. For simple needs, that can be exactly right.

Why is a custom build more expensive than a SaaS tool?

Because someone’s building it around your actual business — your call flows, your software, your edge cases, your compliance. That engineering is real work. It’s also what makes the agent good enough that callers don’t hang up and appointments land in your calendar automatically.

Will an AI voice agent pay for itself?

Usually it comes down to the calls you’re currently missing. If you’re losing jobs to a phone that rings out, the math gets favorable fast — we break it down in the cost of missed calls for a small business. Run your own numbers against the ranges above.

If you want a real number instead of a range, that’s what our discovery is for. Book a free intro call and we’ll map your calls to a fixed, written quote — no cold estimate, no surprises.