AI for a small business fits three buckets. Two are real.
Everything being sold to you as “AI for business” fits one of three buckets: assistant tools your people use at their own keyboards; automation wired into your order flow, your books, and your inbox; and enterprise platforms built for someone with a thousand employees. If you own a 5-to-50-person small business, the first bucket is worth adopting this week, the second is where the money is, and the third you can ignore without falling behind anyone.
That map is the part the hype never gives you. “Use AI” is advice the way “use software” is advice. The useful question is which bucket a given pitch belongs to — so the rest of this article is a sorting guide, plus the one piece of arithmetic that tells you what bucket two is worth in your own numbers.
Put every pitch in a column first.
A vendor, a peer at a trade meeting, a demo that went around the family group chat — whoever brings you the next AI idea, place it in a column before you spend an hour on it. The column decides what it should cost you and who should be involved.
| Assistant tools | Wired-in automation | Enterprise platforms | |
|---|---|---|---|
| What it is | A subscription your team types into | Software connected to your order system, your books, your inbox | A platform with a sales team attached |
| What it costs | A monthly subscription per person | A scoped build, priced against the hours it returns | Enterprise money, enterprise timelines |
| What it saves | Minutes per person, per day | Recurring hours off the wage line | Nothing, at your headcount |
| Who does the work | Anyone on the team, this week | An engineer, connecting systems you already run | An implementation team, over quarters |
| Your move | Adopt it. Skip the consultant. | Measure first. Build the biggest one. | Ignore until you're a different company. |
Assistant tools: adopt them this week. Skip the consultant.
Drafting the awkward customer email. Summarizing the forty-page contract before the call with your lawyer. Turning a rambling meeting into a task list. First passes at job posts, proposals, and product descriptions. General-purpose AI assistants are genuinely good at this work, they cost subscription money, and your team can start using them before Friday with no project plan at all.
The ceiling is as real as the value. Assistant tools stop at each person's keyboard: they make an individual faster at a task, but they don't move work between people or systems. The order still travels between systems by hand, and the numbers for the weekly meeting still come out of somebody's copy-and-paste hour. Faster typing doesn't fix a slow pipeline — minutes per person is worth collecting, it just isn't the hours-per-week problem you were hoping AI would solve.
And a warning about this bucket: it does not need a consultant. If someone quotes you real money to “roll out” chat tools — a workshop, an adoption program, a readiness assessment — keep the money. Buy a few subscriptions, let your team use them on real work for a month, and keep whatever they're still opening in week five.
Wired-in automation: run the arithmetic before anything else.
The second bucket is different in kind, not degree. This is software joined to the tools your operation actually lives in — it can take an inquiry to a booked job, a delivery to a clean invoice, a quiet account to a paid one — and it's the only kind of AI whose value you can read off your own wage bill. Take any task your team does by hand: count the hours it eats in a week across everyone who touches it, multiply by what an hour of their time really costs you, then multiply by 52. That's the annual bill for doing it manually.
Owners are consistently surprised by the size of that figure. The ROI calculator runs the math on three sliders, no signup: a 22-person team spending 8 hours a week on manual work at a $65 loaded rate is $594,880 a year. Your numbers will differ — the point is that you can know yours by Friday, and once you do, every AI pitch gets easy to judge. It either attacks that figure or it doesn't.
Wiring automation in is also less invasive than owners fear. It sits on top of what you already run — anything with an API, a database, or even an export can be joined up, and nothing you rely on gets replaced. That's the shape of my workflow automation work: real code on your infrastructure, written down, and yours to keep when it's done.
The failure mode in this bucket isn't bad technology — it's building the wrong thing first. The fix is boring: measure before you build. The operations audit is that measurement done for you — one week, $4,500 fixed, ending in the short list of builds worth doing, in order, with the savings math for each. The fee counts toward the first build you commission — and if the week doesn't find savings bigger than its own price, it costs you nothing.
What small business owners can ignore. Guilt-free.
The loudest part of the AI industry isn't talking to you, and pretending otherwise is how small companies overspend. Three things you can decline without reading the second email:
- Enterprise AI platforms. The pricing, the rollout timeline, and the governance features were all designed for a company with a thousand seats to spread them across. At 5 to 50 people, you'd be paying for machinery you will never turn on.
- Training your own model. Your edge is your process knowledge and your customer relationships, not a model. The general-purpose models keep improving without you funding the research — the value at your size is in what they're connected to, not in owning one.
- The AI transformation roadmap. A paid vision of your AI future is a purchase that produces a meeting. What changes a company your size is one working system aimed at your biggest manual cost — then a second one.
One more thing may belong on that list, at least for now: outside builders, me included. If the wage-bill math comes back small on every task, bucket one is your entire AI story this year — a few subscriptions, nothing wired in. That isn't falling behind; it's the sorting guide doing its job, and it just saved you the price of a build you didn't need. File the list and check it against next year's volume.
Keep reading: if bucket two is where your math points, What to automate first is the framework for choosing the build, and What AI automation actually costs puts real prices on every way of working with me.