If you're weighing an answering service against an AI voice agent, Smith.ai is probably on your list. It should be. They've been answering phones for small businesses since 2015, they're good at it, and for a certain kind of business they're the right call — over us.
So let's do this the honest way. We'll tell you what each model is actually good at, and we'll tell you plainly where you should buy Smith.ai instead of deploying an agent like ours.
What Smith.ai actually is
Smith.ai's own site describes two front desks: a human virtual receptionist service, and an AI receptionist with a human backup — the AI answers first, and hands off to one of their North-America-based agents when a call gets complex. Both run outside your phone system. You forward your business line to Smith.ai, and their people (or their AI) pick up as your company.
That forwarding model is the thing to understand, because it's where the two approaches genuinely diverge — not the price.
Neither of us posts a price you can trust — for the same reason
We'll be straight about our own side first: we don't publish a flat price either. A 15-seat law office and a 200-seat clinic are different deployments, so we scope each one and quote it. Smith.ai's pricing pages are, as of this writing, a contact form too. Anyone quoting you an exact Smith.ai number from a comparison blog is reading an old cache; price it out with Smith.ai directly. [Update, August 2026: this is no longer true — Smith.ai now publishes per-call tiers. See the current numbers below.]
What you can compare is the meter. Smith.ai has long billed per call — you pay by the conversation, and overages stack as volume grows. A deployed agent has no meter of that kind: it runs on telephony and model accounts you hold in your own name, so what you pay the vendors is what they charge, and the implementation is scoped work rather than a subscription tier. Below a certain call volume, per-call billing is cheaper. Above it, the meter is the thing that surprises you at the end of the month. Neither is dishonest; they're built for different volumes.
The honest axis isn't price. It's what each model is good at.
| Human answering service (Smith.ai) | Teams-native AI agent (Aria) | |
|---|---|---|
| A hard, emotional, or ambiguous call | Strong. A trained human reads the room and improvises. | Handles the routine reliably; escalates the rest to your team. |
| 3am, holidays, a snow day | Covered, at the same per-call rate. | Identical quality at 3am as at 3pm, no premium. |
| 40 calls land in the same 10 minutes | Callers queue for the next available agent. | Answers all of them at once — no hold. |
| Where it runs | Their call center. You forward your line out. | Inside your Microsoft Teams Phone or Webex Calling — no forwarding. |
| Writes to your CRM / books the meeting | Yes, via their integrations. | Yes, natively, on the tools you already run. |
When you should buy Smith.ai instead
We mean this. There's a real profile where a human answering service is the better purchase, and if it's you, go sign up with them:
- Every call is high-stakes intake and low in volume. A personal-injury firm taking twenty calls a day, where the first thirty seconds of empathy decides whether someone becomes a client — that's a human's job, not an agent's. The economics of per-call pricing also favor you at that volume.
- You don't run Teams Phone or Webex Calling. Aria's whole advantage is living inside a cloud phone system you already own. If you're on a plain PSTN line or an old PBX, a forward-out answering service is simply less to change.
- You specifically want a person on every call. Some brands want a human voice as a matter of principle. That's a legitimate choice, and no AI — ours included — is the thing you want.
When a Teams-native agent fits
The picture flips when your reality looks like this:
- You already run Microsoft Teams Phone or Webex Calling. Then the receptionist should live where your calls already live — not forward out to a third party's call center and hand you back a transcript. A voice agent answers inside the phone system you already own, on the numbers you already own.
- Volume is higher, or spiky. A marketing push, a product recall, a seasonal rush — an agent answers the 200th simultaneous call the same way it answered the first. A per-call meter, and a queue, both get uncomfortable here.
- You want it operated, not just switched on. Leverage Automated builds the call flows, wires the CRM and calendar, and watches it in production. That is labour, not a login — the platforms underneath stay in your name and you pay those vendors directly.
Update — August 2026: both prices came out from behind the contact form
When this published we told you neither side posts a number you can trust. That has changed on both sides — and one of the changes contradicts something we wrote above, so let's start there.
Smith.ai publishes real tiers now. We said their pricing page was a contact form. It isn't anymore. Smith.ai's pricing page currently lists Starter at $300/month for 30 calls, Basic at $810 for 90, and Pro at $2,100 for 300 — with per-call overages of $11.50, $10.50 and $8.50 as you move up the tiers. It's month-to-month with no long-term contract, and a 12-month subscription takes 10% off.
That's enough to do arithmetic with, which is the point. Starter works out to $10 a call before a single overage; Pro is $7. Both are knowable in advance now. What hasn't changed is the structure — you are still buying calls, and the overage rate is still the number that decides your bad months.
One correction to our own record while we're here: we also wrote that Smith.ai has "long billed per call." That's still accurate, and the published tiers confirm it rather than contradicting it. What we got wrong was the availability of the numbers, not the model.
Microsoft's native agent has a price. It just isn't in dollars.
The bigger news is on the Microsoft side, because the question this post left open — what does the native option actually cost? — now has a partial answer.
Microsoft's licensing reference is explicit: "During the Frontier Public Preview period: No extra Teams Phone licensing is required for the base Teams Phone Agent features." You still need at least one Teams Phone license and a free resource-account license — which you already have if you run Teams Phone. So the base feature is, right now, free to try.
Then it gets hard to budget. Voice work through Copilot Studio is metered in Copilot Credits per minute: 10 credits a minute for classic voice, 35 for GenAI voice, 75 for premium GenAI voice.
Microsoft does not publish what a Copilot Credit costs in dollars on any page we could load. The billing pages state the rates in credits and point at a licensing-guide PDF; the Copilot Studio pricing URL returns a 404. We're not going to convert those numbers for you, because any figure we produced would be arithmetic resting on an assumption and you'd have no way to see the seam. If a vendor quotes you a dollar-per-minute for Teams Phone Agent, the useful question is where they got the credit price.
And the custom-agent path is still gated. Microsoft's own documentation calls the Copilot Studio integration "a preview feature," says "preview features aren't meant for production use," and states that "tenants must be explicitly enabled… Contact Microsoft support to request enablement." Only basic voice agents are supported; real-time voice agents aren't. Generative orchestration is "currently not supported", with Microsoft warning it "might introduce performance issues and heavy latency, such as 5-6 second delays." Publishing the agent also requires setting authentication to No authentication.
So the honest read on the native option, today: genuinely free to pilot if you already run Teams Phone, genuinely not yet something you can budget for or put in front of customers at scale. That's not a knock on Microsoft — it's what "preview" means, stated plainly. It does mean that anyone telling you to wait for the native agent is telling you to wait for a price that hasn't been published and a GA date that hasn't been announced.
The one-line version
Buy a human answering service when a person is the product — low volume, high empathy, no cloud phone system to build on. Deploy an agent when coverage and integration are the product — you're on Teams or Webex, volume is real, and you'd rather the receptionist live inside your stack than forward out of it.
Not sure which one you are? The fastest way to find out is to hear the AI side yourself. Call our own main line — (206) 578-5242 — and try to trip Aria up. If a human answering service is still the better fit for you afterward, we'll tell you so.
Related reading: We answer our own phone with Aria — what a deployed voice agent actually sounds like on a real business line, and the RingCentral vs. Webex AI receptionist breakdown — how the phone-vendor bundles price out.
Related service: AI integration — a voice agent built into the phone system you already own, on accounts you hold.