Where should an SME start with AI?

Three in the afternoon, and the owner is still personally editing a quotation. For a lot of SMEs, that's just Tuesday.

It isn't that they haven't heard of AI. It's that there are too many tools, the talk is too abstract, and every success story reads like an advertisement — the more you research, the less you know where to begin. So it usually ends one of two ways: sit tight and keep editing quotations, or buy a stack of subscriptions and discover, three months on, that there are more accounts than users.

Our advice has always been unglamorous: don't look for the most advanced thing, look for the most convenient thing. There are three small jobs almost every company can start with. Each costs little, pays back fast, and doesn't hurt much if it fails.

Incidentally, "start by doing it yourself" has a lower bar than most owners assume. An independent developer at a recent talk did the arithmetic: a month of one employee's cost buys several top-tier AI subscriptions. He built a phone app to manage his own bank cards in about two days — most of which was spent waiting for the AI to finish. The point isn't that owners should write software. It's that the person who knows the business best doesn't have to wait for a technical team to get everything ready; they can pick up the most annoying task and try it themselves.

One: the repetitive paperwork

Every company feeds a pile of highly repetitive writing: quotations, contract first drafts, customer emails, meeting minutes, data entry. The pile has three things in common — it's large, it's mechanical, and nobody enjoys it. Which happens to be exactly what today's AI is best at.

Keep the method simple. Pick one document that shows up at least ten times a week. Dig out the three to five best past examples and make them the template. Let AI draft; a person reviews and signs off. Run it for two weeks, count the hours saved, then decide whether to widen it. A step this small is often enough to change how the team talks about AI — from "I hear it's impressive" to "it actually works".

Two: the dormant know-how

An SME's most expensive asset is usually locked inside a few veterans' heads: how far a price can stretch, which client needs a soft touch, which supplier will move if pushed, where the company came to grief in some earlier year. When those people leave, all of it leaves with them. When new people join, everything depends on whether the mentor feels like teaching — and how much they remember to pass on.

Gather the experience scattered across chat logs, emails and folders into an internal knowledge base, put AI-powered search and Q&A on top, and most of a newcomer's questions now have an old hand on call at any hour. The value here isn't technical. It's that "only she knows" becomes "the whole company knows" — for the first time, the firm owns its experience instead of renting it from employees.

Three: the late reports

Plenty of owners see last month's numbers in the middle of this month, so every decision runs half a beat behind. Consolidating the data scattered across POS, inventory and bank records into a one-page weekly report is the classic build-once, benefit-for-years project. AI's job inside it is modest: translate the numbers into plain language — where the anomaly is, which way the trend leans, what to watch next week. The owner's instincts still matter. They just stop working alone.

Three field notes

Doing these three, a few lessons are worth keeping in mind. Work out how a step ought to run before choosing any software — wiring AI into a chaotic process just produces faster chaos. Give each experiment two to four weeks: scale what works, kill what doesn't, and don't get attached. And the easiest one to forget: the saved hours need a destination. Steer them towards customers, product and the decisions that actually need a brain, and the returns arrive; otherwise you've merely traded one shape of busyness for another.

Transformation doesn't begin with a blueprint. It begins with the one small thing you can start on Monday morning.