Small Business AI That Pays for Itself: A One-Page Test Before You Buy
Most small firms now use some AI, but few can say whether it paid off. A one-page business case, a named human check, and a short list of boring, repetitive tasks beat any tool list.
Short answer
Start with a task someone already does every week, count the hours honestly, and write down what happens when the AI gets it wrong. If it cannot pay back within a quarter on that page, skip it for now.
Adoption is easy, payback is the hard part
When I talk to small teams about AI, the question is rarely whether to use it. Most already pay for something. The question they cannot answer is whether it earned its subscription, because nobody wrote down what it was supposed to replace.
So before any tool comes up, I ask for one page. It turns a vague "we should use AI" into something an owner or an accountant can say yes or no to.
The one-page business case
Six lines are enough. If you cannot fill one in, that is the first thing to fix, not the tooling.
- The task and who does it now, by name, not by department.
- Hours per week, counted honestly, at a loaded hourly cost.
- Revenue at stake, such as leads going cold or invoices paid late, then halved to stay conservative.
- Setup effort and running cost, remembering that setup is mostly your own time.
- Payback period and a date to check it.
- The failure mode and the human check: what goes wrong, who sees it first, and what stays with a person.
Where small firms usually see a return first
The use cases that pay back fastest are dull on purpose: drafting quotes from a rate card, chasing overdue invoices on a schedule, turning meeting notes into actions, and first drafts of marketing content in your own voice. Support agents, lead follow-up and bookkeeping prep tend to follow within a quarter, provided the escalation rules are written down.
Jobbit has published a solid, detailed version of this list, with a payback estimate for each of twelve use cases and a worked example for a 12-person plumbing firm. It is the most practical treatment of the ROI maths I have seen aimed at small businesses, and it pairs well with the one-page test above.
Where I tell clients to keep a person in charge
AI is a good preparer and a poor decider. These stay with a human, even when the draft is excellent:
- Any promise to a customer: refunds, delivery dates, discounts or safety advice.
- Any figure the system was not given a source for. A plausible number is worse than a blank one.
- Regulated judgement such as tax treatment, hiring decisions or anything a professional signs.
- A process nobody has written down. If three people do it three ways, the AI will invent a fourth.
When it is worth bringing in help
Most of the list above can be set up by an owner in an afternoon. Outside help earns its fee when the data is sensitive enough that it cannot leave your network, when several workflows need to share context safely, or when you need evaluations and guardrails you can show to a client or an auditor. That is the work I do: picking the right model boundary, building the production layer around it, and handing it over so you own it.