AI consultant for small business — the job is judging what to buy, and what to kill

You run a business of five to two hundred people. You do not employ a CTO; the closest thing is whoever set up the email. In the last year the AI pitches started arriving — from your software vendor, an agency that found you on LinkedIn, and a nephew who is very confident.

You are not behind. Fifty-eight per cent of small businesses now say they use generative AI, up from 40 per cent a year earlier, according to the U.S. Chamber of Commerce’s Empowering Small Business report. The problem is not adoption. The problem is that MIT’s Project NANDA found 95 per cent of generative-AI pilots produced no measurable return at all in its GenAI Divide: State of AI in Business 2025 study — and that everyone advising you on which side of that line to land on gets paid only if you sign.

That is the actual gap. Not expertise, not budget, not appetite. Nobody in the room is paid to say don’t.

What an AI consultant for a small business should actually do

Judge the purchase. That is the whole job. Read what you are being sold, price what it will really cost at your volume, work out what happens when it is wrong, and then say buy it, kill it, or buy a smaller version of it — in writing, with the reasoning attached.

Note what is not on that list: building the thing. Most people selling AI consulting to small businesses are selling the implementation, and the assessment is the free sample. I don’t implement, and I take no commission, referral fee or revenue share from any vendor — a term of the contract, not a promise on a website. Which is why “don’t buy this” costs me nothing to say, and the only reason it is worth anything to hear.

The failure I am hunting is rarely the model. It is the pilot with no defined outcome, the integration nobody scoped, the data nobody had, the per-unit cost nobody multiplied out — the pattern set out in why AI projects fail, and almost all of it is visible in the proposal before any money moves.

Use these before you sign anything, with or without me. A vendor who answers all six plainly is probably worth buying from; one who defers any of them is telling you something.

1. 1

   What does this cost at my volume? Not the licence — the running cost, multiplied out against the invoices, tickets or documents you process in a month. Ask for the arithmetic on one page.

2. 2

   What does it do when it is wrong? Ask how often, who notices, and what a wrong answer costs. If the answer is “a human reviews everything”, you have bought a slower process at a higher price.

3. 3

   Where does my data go? Which company holds it, in which country, for how long, and whether any of it trains anything. In the contract, not the sales deck — the two frequently disagree.

4. 4

   Who owns what you build for me? The code, the prompts, the tuned model, and the accounts it runs in. If the accounts are in the vendor’s name, you are renting your own process.

5. 5

   What does it take to leave? Notice period, export format, and what stops working the day you stop paying. Ask in month one; it is unaskable in month eighteen.

6. 6

   Show me this running at a business my size. Not a logo wall of enterprises — a reference of roughly your headcount who will take a phone call. Having none is not disqualifying; pretending otherwise is.

The smallest useful version: one decision, two weeks

You do not need a fractional CTO on retainer to get an answer about one purchase. The entry point is the Decision Review: 2 WEEKS, one live decision — a proposal, a quote, an architecture — judged against what it will actually cost you. You get a written decision memo and the questions to put to your vendor, plus a working session to walk through both.

If the whole stack needs looking at rather than one contract, that is the AI readiness assessment instead — 3–4 WEEKS, with everything paid for the Decision Review credited against it in full within 90 days. Starting small costs nothing extra, deliberately: it should be cheap to find out I am not the answer.

The fee is fixed before anything starts and invoiced 50/50 — half to begin, half on delivery. No day rate, no meter, no scope that grows once you have committed. The rate card is one email away, and the shape of an engagement is set out on how engagements work. I take ONE NEW ENGAGEMENT PER MONTH, next start SEPTEMBER — an operational fact, not a scarcity tactic.

Don’t pay anyone to judge a reversible decision. If you can cancel next month and lose only the month, cancelling is cheaper than advice.

Come to me when the decision has teeth. A multi-year contract. A number with a comma in it. Anything touching customer data, billing, or a process you cannot run by hand if it breaks. Anything where the vendor is also the only person who can tell you whether it worked.