You asked three vendors for "something that answers my phones and follows up with leads," and got back three numbers with nothing in common: $99 a month, $650 a month, and a $22,000 proposal. Nobody explained why they're so far apart. Here's what actually drives AI agent pricing, in plain English, so you can tell whether a quote is fair before you sign anything.
Small businesses paying for an AI agent system usually land somewhere in the low hundreds to low thousands of dollars a month for an off-the-shelf or lightly-customized setup — with a fully custom-built system running into five figures upfront instead. The huge range exists because "AI agent" pricing actually hides three different models: flat monthly (per-seat), usage-based (per call or task), and one-time custom build. Once you know which one a quote is using, and how it maps to your actual call and message volume, comparing vendors gets a lot less confusing.
"AI agent" is doing a lot of work as a phrase. It's a little like asking three people to quote you "a vehicle" — one comes back with a scooter, one with a sedan, and one with a delivery truck, and all three are technically answering the question. The word doesn't tell you how much the thing can handle, how it's billed, or what happens when your busiest month is triple your slowest one. That's why an owner can get a $99/month quote and a $22,000 quote for what sounds, on paper, like the same request.
The fix isn't finding the "real" price — there isn't one. It's learning to recognize which of three pricing models each quote is actually using, so you're comparing apples to apples instead of a subscription to a construction bid.
1. Flat monthly, also called per-seat pricing. Per-seat pricing means you pay one fixed amount per month, similar to a Netflix subscription, regardless of how busy the system actually is that month. You're usually paying for a tier — a cap on how many conversations, calls, or tasks are included — and if you go over, you either get bumped to a higher tier or charged an overage fee. This is the easiest model to budget for, and the easiest one to get an unpleasant surprise from if your busy season blows past the cap nobody flagged for you upfront.
2. Usage-based pricing. Usage-based pricing charges you per action — per phone call answered, per text sent, per booking handled — so your bill moves with your actual business volume. A slow month costs less. A busy month, or a marketing push that triples your inbound calls, costs more. This model is fairer in theory, since you're paying for what you use, but it's harder to budget for if your volume is unpredictable, and it rewards vendors who don't explain clearly what counts as one "unit" of usage.
3. One-time custom build, plus a smaller ongoing fee. Here you pay a developer or agency a large upfront amount — often five figures — to build a system shaped exactly around your specific workflow, followed by a smaller monthly fee to keep it hosted and maintained. This makes sense when your process is genuinely unusual and an off-the-shelf tool can't bend to fit it. It rarely makes sense for a first attempt at automating one task, which is where most small businesses should actually start.
Take the free 2-minute readiness assessment
Inside any of those three models, four things move the number: how many conversations or tasks the system handles per month, how many other tools it needs to connect to (your calendar, your CRM — the software that tracks your customers and jobs — and your payment processor), how much human review and oversight you want built in, and how far off-the-shelf the workflow is from what you actually do. A system that just answers common questions and logs a message is cheap. A system that reads an incoming request, checks your calendar, quotes a price, and books the job without a human touching it is a different, more expensive thing — even though a vendor might describe both as "an AI agent" on their homepage.
This pricing confusion is showing up more, not less, precisely because adoption is moving fast. The U.S. Chamber of Commerce's 2025 small business survey — based on responses from roughly 3,870 small businesses nationwide — found that 58% of small businesses now report using generative AI, up from 40% the year before.1 That's a lot of business owners shopping for pricing they don't have a frame of reference for yet, which is exactly the gap vendors' inconsistent pricing language falls into.
Say two small service businesses, each handling roughly 300 customer calls and texts a month in an average month, are choosing between the same two options. Business A picks a flat $149/month plan capped at 250 interactions. In a normal month they're fine, but their December volume triples to 900 as holiday requests pile up, and the overage fees push that month's bill to $610 — more than four times the sticker price they budgeted for. Business B picks a usage-based plan at $0.35 per handled interaction. Their bill runs about $105 in a slow month and $315 in their busiest month, averaging closer to $170/month across the year — lower than Business A's flat plan once you count the December spike, precisely because their volume is uneven rather than steady. Neither model is "wrong." The mismatch is what costs money — a flat plan sized for average volume, on a business with spiky volume.
Not necessarily, and it's worth being honest about why. A very low flat price can mean a low usage cap with expensive overage fees waiting past it. A very low per-unit usage price can mean a narrow definition of "one unit" — some vendors count an incoming text and your reply as two separate billable actions, which doubles your real cost without changing the number on the pricing page. And a large custom-build number isn't automatically a ripoff either — sometimes it genuinely reflects work that an off-the-shelf tool can't do. As one small business owner might put it, describing a realistic but illustrative scenario: "I picked the $79 plan because it was the cheapest one I found. Three months later my bill was $310, because nobody told me a customer's text and my agent's reply back both counted as separate billable messages. It wasn't a scam exactly — it just was never explained in a way I could compare against the other quote sitting in my inbox." The fix isn't distrust of every vendor. It's asking, before you sign anything, exactly what counts as one billable unit, and what a realistic busy-month bill looks like — not just the advertised starting price.
We price around your actual workflow instead of a generic seat count, and we walk through what a realistic month — including your busiest month, not just an average one — would cost before you commit to anything, using the same kind of comparison laid out above. The growth case for paying for this at all is straightforward: the hours a system frees up from your front desk go toward higher-value work, and the calls or messages it stops from going unanswered turn into booked jobs you'd otherwise have lost, which is usually where the real return shows up, not just in the monthly bill itself.
We also back this with a real guarantee, not a vague one: try Unmanually for 60 days, and if it isn't saving your business real time, whatever's left of your prepaid balance converts to account credit. That's not a cash refund on usage you've already consumed, since that reflects real infrastructure cost already spent, but it does mean testing the pricing against your real volume for two months isn't money you're simply gambling.
Not ready to get a quote yet? That's fine — leave your email on our presale waitlist and we'll let you know the moment founding-member presale pricing opens, before it's available to everyone else this October.
1. U.S. Chamber of Commerce, "U.S. Chamber's Latest 'Empowering Small Business' Report Shows Majority of Businesses in All 50 States Are Embracing AI," 2025 (survey of roughly 3,870 U.S. small businesses with fewer than 250 employees, conducted June 6–26, 2025; figures describe generative AI usage broadly, not AI agent pricing specifically).
The full pillar guide this article belongs to.
The build-it-yourself side of the same cost question — when a five-figure custom system is actually worth it.
Once you know what it costs, here's how to tell if that price is actually paying for itself.
Start here if you're not yet sure what you'd even be pricing out.
It depends on which of three pricing models you're being quoted. Flat monthly (per-seat) tools for a small business typically run somewhere in the low hundreds of dollars a month. Usage-based tools charge per call, message, or task, so the bill moves with how busy you are. A fully custom-built system usually means a larger one-time development cost plus a smaller ongoing hosting and support fee. The number that matters isn't the sticker price alone — it's the sticker price compared against what the model actually predicts you'll use.
Per-seat pricing charges a flat amount per month, similar to a Netflix subscription, no matter how much or how little the system actually does that month. Usage-based pricing charges per action — per call answered, per message sent, per task completed — so a slow month costs less and a busy month costs more. Neither is automatically better; the right one depends on how steady or spiky your business's call and message volume actually is.
Take our free 2-minute readiness assessment — it looks at your actual call and message volume and tells you honestly what to automate first, and roughly what that would run.
Take the 2-min readiness assessment