What AI adoption actually looks like in a Welsh SME

AI adoption in a Welsh small business rarely starts with a plan. It starts with one person pasting a client email into ChatGPT because the reply is overdue, finding the draft good enough to send with two changes, and doing the same thing again the next morning. Almost every business that has started did it this way: one task, decided by one person, without a meeting about it. There is also more of it about than the mood music suggests. Welsh Government’s own figures put 49% of Welsh businesses as not currently using AI at all, against 56% across the UK, so Wales sits slightly ahead of the UK average rather than behind it.

What actually happens first

The first real use is almost always words, not automation. A dental receptionist drafting a reply to a nervous patient. A council clerk turning scrawled meeting notes into something publishable. A recruiter writing three versions of a job advert instead of labouring over one. None of it requires touching a business’s systems, which is exactly why it spreads so quickly and so quietly: nobody has to ask IT, nobody has to sign anything, and the only cost is a subscription most people are already paying for something else. A handful of businesses go further early on, using AI to summarise a set of figures or draft a first pass at social content, but the pattern holds: it starts as one person’s shortcut, not a company decision. It’s also where the real risk sits, and it’s the part that usually goes unexamined: information pasted into a public AI tool has left the business, and anything involving customer, patient or staff data needs a decision about which tools are approved before it becomes a habit, not after.

Where it stalls

Where it stalls is just as consistent. AI Cymru: Shaping a Smarter, Fairer, More Prosperous Wales, the Welsh Government’s first national AI strategy, published in November 2025, names the same three barriers we hear directly from businesses across Wales: a lack of expertise, cost, and difficulty identifying a genuine use case. The third one does most of the damage. A business owner who has used a chatbot to write a social media post has no obvious next step, because “write better copy” and “run my invoicing” feel like different problems, and they are. Adoption plateaus exactly there, at one useful trick repeated, with nothing built on top of it.

What it actually costs

The genuine cost is rarely money. Most of what a small business tries first is free or a few pounds a month, and the tools themselves aren’t the constraint. The real cost is time spent learning what not to trust it with: the invoice summarised wrong, the review reply that sounded too corporate, the client email answered confidently and incorrectly. That learning has to happen somewhere, and for most Welsh SMEs it happens on live work rather than in a training session, because nobody has budgeted for a training session. It costs a business an afternoon here and there, and a fair amount of second-guessing, before it costs anything on an invoice.

The pattern across sectors

We see this pattern across every sector we work in, from dental practices and community councils to self-storage and recruitment. It rarely arrives as a single project. It arrives as one person’s workaround that either spreads through the team or quietly stops. The businesses that get further tend to have one thing in common: somebody decided, even informally, what the tool is and isn’t allowed to touch. A dental practice happy to let AI draft appointment reminders will still want a person checking anything that touches patient records. A community council using it to tidy up minutes won’t let it near a planning decision. That line, drawn early, tends to be what separates a business that keeps using AI from one that tried it once and quietly gave up.

For councils and third-sector organisations there’s an extra layer the Welsh Government strategy addresses directly: a stated commitment to multilingual AI tools and to making Welsh-language datasets available to developers, so a system trained mostly on English performs properly in Welsh too. That matters practically. A community council publishing bilingual minutes, or an advice service answering a query in Welsh, can’t use a tool that only works properly in English and call it done. It’s one more reason adoption looks different depending on who you are: a private dental practice can experiment freely with an English-only tool; a public body serving Wales bilingually generally can’t.

The other shift, happening at the same time

There’s a second, less obvious change happening alongside all this. Google’s AI answers and ChatGPT increasingly answer a customer’s question directly rather than sending them to a website. In our own work we now regularly see a search for a local dental practice or a self-storage unit answered by an AI summary before a single link is clicked. That changes what a business’s own website needs to do: it needs to be written clearly enough, and structured plainly enough, that an AI system can read it and cite it accurately. Adopting AI internally and being findable by AI externally are two separate problems, and most Welsh SMEs are only just starting on the first one.

The honest starting point

None of this is really about the technology. Wales’s own government data shows just under half of businesses here still aren’t using AI at all, and plenty of the ones that are have stumbled into it rather than planned it. That’s not a bad place to be. It just means the honest starting point for most Welsh SMEs is smaller and slower than the pitch decks suggest: one task, one person, a line drawn around what the tool doesn’t get to touch, and a willingness to be wrong a few times before it’s actually useful.

---

Sources

• AI Cymru: Shaping a Smarter, Fairer, More Prosperous Wales (GOV.WALES, published 18 November 2025). Fetched directly for the 49%/56% business-adoption figures, the three named adoption barriers (expertise, cost, unclear use cases), the Business Wales/Hartree Centre support commitment, and the Welsh-language AI/dataset commitment.
• Internal, verified against source: src/data/clients.ts in the webjects-website repository. The client sector categories referenced (dental practices, community councils, self-storage, recruitment) are drawn from Webjects’ own live, public client roster, not invented.