Every SME owner I meet has heard the same pitch by now: AI is transforming business, you need a strategy, and if you don't move fast you'll be left behind. Most of it is noise. The businesses actually getting value from AI right now are not the ones with the fanciest strategy decks — they are the ones who picked one or two genuinely annoying admin problems and automated them properly.
This article is the version of that advice I give clients in the room, without the sales pitch attached.
Start With the Work, Not the Technology
The biggest mistake I see is starting with a tool — "should we use ChatGPT, or Copilot, or build something custom?" — before anyone has written down what the actual problem is. Flip it around. Walk through a normal week in the business and write down every task that is repetitive, rules-based, and takes longer than it should. Quoting. Chasing invoices. Answering the same five customer questions. Writing job descriptions. Summarising site visit notes. Sorting incoming enquiries by urgency.
None of that list needs to mention AI. It's just a list of friction. Once you have it, AI becomes one possible answer among several — sometimes the right one, sometimes not. A lot of "AI problems" turn out to be process problems that a simple form or a shared spreadsheet would fix just as well, for free.
The 80/20 Test
For each item on that friction list, ask two questions. First: does this task follow a pattern, or does it need real judgement every time? AI is excellent at pattern-following — drafting a response, extracting data from a document, summarising a call — and much weaker at genuine judgement calls involving risk, relationships or nuance. Second: what does it cost you right now, in hours or in missed opportunities, if this stays exactly as it is for another year?
Tasks that are pattern-based and expensive to leave alone are your starting point. Everything else can wait. Most businesses I work with find two or three tasks meet that bar — not twenty.
Pilot Small, on Real Work
Once you have a candidate, resist the urge to plan a company-wide rollout before you've tried it once. Pick one process, one team member, and one week. Use an existing tool if one does the job — many businesses don't need custom development at all, just the right configuration of tools they may already be paying for. Measure the time saved or the quality improvement honestly, including the time spent checking the AI's output, because that check is real and it counts.
If it works, you have evidence rather than a hunch, and a template for how to extend it. If it doesn't, you've lost a week rather than a quarter and a budget.
Where Custom Automation Earns Its Keep
Off-the-shelf AI tools cover a lot of ground, but there is a category of problem where a bespoke workflow — connecting your CRM, your inbox, and an AI model together with proper rules — clearly outperforms a generic chatbot subscription. Anything that needs to read data from one of your existing systems, apply judgement, and write the result somewhere else usually falls here: lead qualification, quote generation from a spec sheet, automated first-draft replies that pull from your actual pricing and policies rather than generic web knowledge.
This is where rapid AI development work tends to pay for itself fastest, because it removes a task entirely rather than just making it slightly quicker to do by hand.
What Good Looks Like After Three Months
If this has gone well, you should be able to point to one or two specific processes that now take meaningfully less time, run with fewer errors, or free up someone senior from work a junior process shouldn't need them for. You should not need a new department, a six-figure software budget, or a rewritten five-year plan. AI strategy for most SMEs is not a transformation programme — it's a short list of well-chosen automations, tested honestly, and expanded only once they've proven themselves.
If you want a second pair of eyes on your own friction list, a free AI audit is the fastest way to get one — no obligation, no jargon, just a straight answer on what's actually worth doing first...