All posts
Operating model

What AI should cost an SMB

Nobody will give you a straight number, so here's ours. What a first project, a full agent, and ongoing ownership actually cost a business under 500 people — and the pricing shapes that should make you walk.

Published November 17, 2026
Reading time 5 min
By Dominic Morgan

Ask what AI costs and you get a shrug and a “depends.” It does depend — but not so much that nobody can give you a number. The vagueness is usually there to keep you in a sales process, and it leaves smart owners assuming the answer is either a few hundred dollars a month or a quarter of a million. It’s neither. So here are real numbers and the reasoning behind them.

The build is the small cost. The one that actually decides your return is ownership.

The three things you might be buying

Most SMB AI work is one of three shapes, and they cost very differently because they are genuinely different amounts of work.

A first project — roughly $5–8K. One clearly-scoped job, taken from idea to working, in weeks not months. A report that assembles itself; a knowledge agent that answers your top questions. This is deliberately small: the point is to prove the thing works on your real inputs before you spend more. If someone wants six figures to find out whether AI can help you at all, that’s a red flag, not a commitment.

A full agent build — roughly $8–20K. A real piece of automation running in your operation, wired into the systems it needs, with the guardrails and handoffs that make it safe to trust. The range is wide because it tracks how many systems it touches and how much it’s allowed to do on its own — a read-only assistant is one thing, an agent that takes actions is another.

Ongoing ownership — roughly $3–6K a month. The part most people forget to price, and the part that decides whether any of it keeps paying off. More on why in a moment.

Why ownership is the number that matters

A build is a one-time cost. An agent is a living thing in your business: the world it operates in changes, your policies change, the questions people ask it change. Left alone, it drifts — slowly answering yesterday’s questions with yesterday’s rules — and the trust you spent months earning evaporates in a week of wrong answers.

Ownership is what stops that: someone reads what it got wrong, closes the gaps, and keeps it honest. It’s the difference between an asset that compounds and a project that quietly rots, which is the whole argument of who owns the agent and products, not projects. If a proposal has a big build number and no ownership line, it’s not cheaper — it’s unfinished.

How to know it’s worth it

Don’t start from the price. Start from what the job costs you today. The reporting that eats a senior person’s Monday, every week, is a real fraction of a real salary — the maths I walked through in the Monday report that eats your week. Put that annual number next to a build plus a year of ownership. Most of the honest cases clear it comfortably, and the ones that don’t, you want to find out before you spend, not after.

The frequency test does most of the work. A daily or weekly job across your whole book adds up fast; a quarterly one rarely pays for the build. If the numbers are close, that’s a signal to pick a different first project, not to talk yourself into this one.

Pricing shapes that should make you walk

Three patterns cost far more than the sticker over time.

Per-seat pricing that scales with headcount. If the tool costs more every time you hire, you’re punished for growing — and the vendor’s incentive is your seat count, not your outcome. Automation that’s worth it should get cheaper per unit of work as you use it more, not more expensive per person.

Open-ended consulting with no defined deliverable. “We’ll embed a team and see” is how a $10K problem becomes a $200K one. You want a fixed scope, a fixed price, and a working thing at the end — then decide whether to do the next one.

A big number before anyone’s seen your data. If the price is set before anyone has looked at your actual reports or tickets, they’re pricing a fantasy. The honest sequence is small paid diagnostic, then a scoped build — roughly what our two-week audit is.

When the answer is “not yet”

Sometimes the right number is zero. If the only jobs on the table are rare, or genuinely bespoke every time, or the data lives on paper, the build won’t clear its cost and you should wait — or fix the input first, which is the argument in your data isn’t ready — start anyway. A firm that tells you to wait when waiting is right is the one worth hiring when it’s time.

If you want to sanity-check where your own numbers land, the readiness scorecard is a five-minute start. And once you’ve decided it’s worth it, the next thing that quietly wastes the money isn’t the price — it’s a tool nobody ends up using, which is where why your team ignored the tool you bought goes next.

Want to talk about this in your business?

Most of what we write comes out of actual client work. If anything here resonates, we'd be happy to dig into the specifics with you.