The biggest AI win in most small businesses is not an AI feature. It is a workflow fix, and this method is built to find that rather than hide it. It takes about two weeks and four hours of the owner’s time, and is written so you can run it yourself or hand it to someone.
The starting point is usually the same: vendors arriving with a product to sell, staff experimenting quietly, a larger customer asking for a policy, and nobody having looked at the whole business and asked where the money actually is.
Week one: understand how the business actually runs
Short conversations, thirty to forty-five minutes each, with the owner and with the people doing the work. Not the org chart; the people who re-key, chase, wait and check.
For each, ask for a walk through a normal day and note every hand-off: where a piece of information moves from one person or system to another. Hand-offs are where time leaks. Watch for the four signs:
- Re-keying: the same data typed into two places.
- Chasing: waiting on someone to respond before work can continue.
- Waiting: a task blocked on a decision or approval.
- Checking: verifying something a system should have verified.
Produce a workflow and systems map: how work moves through the business today, which tools it touches, and where the friction is. One page per major workflow. Plain language; the owner will read it.
Week two: score, rank and plan
List every candidate for software or AI that came up, including the boring ones. Score each on two axes.
Payoff. Hours saved per week, multiplied by who is doing them; errors avoided and what they cost; revenue unblocked. Be honest about the number. “Saves the owner two hours a week” is a real figure; “transforms the business” is not.
Effort. Whether it can be done with a tool the business already pays for, a configuration change, an off-the-shelf product, or custom software. Whether the data it needs exists and is clean. Whether the people who would use it are ready to.
Rank by payoff over effort. The boring wins rank first almost every time: automating the quote follow-up, one source of truth for the customer list, a form that replaces the email thread. Say so plainly.
Then check readiness across four things:
- Data: does it exist, is it trustworthy, who owns it.
- Systems: can they be connected, or are they closed.
- Skills: who would run this, and do they want to.
- Risk: what the business must not get wrong (money, personal data, a regulated obligation).
Anything that fails readiness moves down the list with a note on what would need fixing first. Then write the roadmap: what to do first, second and third, with a realistic cost and timeline for each and the score explained in plain English.
Wrap-up: a walkthrough and a report
One hour with the owner, walking through the findings. Then the written report, in four parts:
- The workflow and systems map
- The scored opportunity list, every candidate, with the reasoning
- The recommended roadmap, first, second, third, costed
- The readiness check: what would need fixing before the roadmap can start
Everything is written for a business owner, not a developer. The test of a good report is that it could be handed to any competent developer or vendor and they could act on it.
Traps
The vendor's problem. Three quotes for "an AI system" and no way to compare them.
The most common starting point. The assessment reframes the question from “which vendor” to “which problem”, and often the answer is a fraction of the cheapest quote.
Confusing adoption with value. Staff using an assistant is not the business getting value from it.
Include a line in the readiness check on how AI is currently used, consistently or not, and whether there is a policy.
Scoring what is exciting.
Score the boring candidates first, before the interesting ones, so the interesting ones have to beat them.
Building before deciding.
The owner does not have to build anything. The value of the report is that the decision has been made on evidence, including the decision to do nothing.
Changelog
- 2026-09-05: first version, released from the assessment method.
- 2026-09-08: the four signs and the readiness check made into lists; traps collapsed; short version added.