For businesses still watching the AI boom from the sidelines, the pressure to catch up can feel increasingly difficult to ignore.

But for Jaymie Thomas, Co-founder of AI Wales, the biggest mistake Welsh SMEs can make is allowing that pressure to stop them from getting started at all.

Official figures suggest Wales is already moving beyond the earliest stages of adoption. According to the Welsh Government’s AI plan, 49% of Welsh businesses reported not currently using AI technologies, compared with 56% across the UK. Among those making use of the technology, machine-learning data processing, large language model text generation and visual content creation were the most commonly reported applications.

Thomas, however, sees a business community moving at very different speeds.

“I think it’s a mix,” he told DIGIT. “Some firms are excited by what’s going on. Some are feeling pressure because they either perceive their competitors are doing something or, you know, they hear about a project someone else has going on and maybe they don’t have that yet.

“I think a lot of leaders can feel guilty, especially if they haven’t started any AI initiatives yet. So there’s always not just fear of being behind, but fear of missing out completely. That guilt can paralyse them from starting at all.”

At the other end of the spectrum, AI adoption may already be happening inside organisations without leaders necessarily having a clear view of it.

“A lot of users’ activity is individual staff using free tools if there hasn’t been a planned business approach,” Thomas said. “If you don’t have an AI policy, your staff are using AI already, whether you know it or not.”

Recent Office for National Statistics data illustrates the wider acceleration in workplace adoption. Around 35% of UK businesses with 10 or more employees reported using at least one AI technology in 2026, up from around 12% in late 2023.

But official business-level adoption figures may only tell part of the story.

Deloitte’s 2026 GenAI Workforce Survey, based on responses from 25,000 UK workers, found that 31% of people using generative AI for work were doing so without their employer’s knowledge. Almost half (46%) were using free-to-use AI tools at work, while 17% were paying for AI tools themselves.

The findings underline the growing “shadow AI” challenge facing organisations as employee adoption moves ahead of formal oversight, policies and approved technology.

Despite the rapid increase, adoption remains relatively shallow. The average number of AI technologies used by adopting businesses has risen only modestly, from around 1.4 in 2023 to 1.6 in 2026.

For Thomas, that is not necessarily a problem.

Rather than beginning with an expensive transformation programme, he believes SMEs can gain more by finding a simple, measurable process and experimenting.

There is also a misconception, he said, that meaningful adoption immediately requires significant investment.

“It’s easy to think that it’s expensive because people hear industry talk of “token maxing” and spending in the region of £500+ per person,” he said. “It doesn’t have to be that.

“People can get started. It might cost them just £20 a month per user. This doesn’t need to be a big investment, at least initially, but just dip your toe, just try some stuff out.”

One simple example could be as straightforward as automatically identifying a one-star customer review, allowing a business to be alerted and respond more quickly, rather than relying on someone to manually monitor feedback.

The principle, he said, is to begin with a small problem where success – or failure – can easily be identified.

That approach is supported by wider research into the barriers holding companies back.

Welsh Government analysis has previously identified lack of AI expertise, cost and difficulties identifying business use cases among the factors delaying adoption in Wales. Meanwhile, Department for Science, Innovation and Technology research found that 71% of UK businesses cited not having identified a use for AI as a barrier to adoption, while 60% pointed to limited AI skills or knowledge. Cost was cited by 23%.

The figures suggest that for many companies, the more fundamental problem is not access to the technology, but understanding where it can genuinely add value.

Start With the Problem, Not the AI

Thomas argues that businesses should resist the temptation to introduce AI simply because the technology is available.

Preparing the underlying organisation first can be just as important.

“Is everything centralised? Is it well categorised?” he said. “Because then the AI can interpret it and surface relevant information in a timely manner.

“If it’s all across thousands of folders, different SharePoint sites, you’ve got to do that consolidation and organisation exercise first so that AI can use it well and obviously so your employees can use it better as well.”

He also warns against making a complex transformation project a company’s first serious encounter with AI.

“I think people are at risk of doing too much too soon,” Thomas said. “The first project they do should be simple because you want it to succeed.

“You don’t want to do something complex, have it fail to deliver benefits, and have that be your experience of AI as a whole; and it should cover a process that staff know very well already so that they can see the differences, the opportunities to improve it.”

That also means accepting that some processes may not need AI at all.

“You can also feel tempted to shoehorn AI into processes where it can cause confusion,” he said. “If it’s working well already, leave it alone. You don’t have to AI all the things.”

When AI is applied effectively, Thomas believes SMEs should initially think less about replacing entire roles and more about removing repetitive elements of them.

“The way to think about this is that AI can replace tasks, not people,” he said. “And then those people are freed up for more meaningful work. Just remove that repetitive admin and drudgery.”

There is evidence that businesses already using AI are seeing those kinds of productivity gains.

DSIT research found 56% of AI-using businesses reported an increase in employee productivity following adoption, while 35% reported no change. However, its immediate effect on the bottom line was much less pronounced: 77% reported no change in revenue.

For businesses taking their first steps, however, experimentation cannot come at the expense of governance.

Thomas recommends establishing an AI policy, defining which tools employees are permitted to use and reviewing how business data is organised and accessed before expanding adoption.

“Have one,” he said of an AI policy. “List those tools that are approved for use, because if you don’t and someone leaks data from the business, you can’t take action against them because you didn’t tell them what was acceptable and what wasn’t.”

The gap between AI adoption and AI governance remains significant.

The UK Government’s latest Cyber Security Breaches Survey found that among businesses using, adopting or considering AI, only 24% had security practices or processes in place specifically to manage AI-related risks. A further 38% planned to introduce them over the following 12 months, while 31% had no plans to do so.

For Thomas, successful adoption therefore involves people and organisational change as much as choosing technology.

That thinking also sits behind the work AI Wales is carrying out with FinTech Wales through its AI Hub.

Rather than prescribing where businesses should deploy AI, Thomas said the aim is to give organisations somewhere to bring their own ideas, problems and questions, before identifying practical opportunities together.

AI Wales also runs training programmes designed to help companies turn that initial experimentation into longer-term adoption. With its AI Pathfinder programme, businesses can work through potential AI opportunities and leave with a 12-month roadmap, while its six-month AI Champions programme supports employees tasked with driving adoption internally.

“This is as much about hearts and minds and concerns as it is understanding which tools do what,” Thomas said.

Welsh firms also currently have access to enhanced public support for AI training.

Under the Welsh Government’s Flexible Skills Programme AI pilot, eligible employers can receive a 75% contribution towards approved AI training costs, with the enhanced scheme currently scheduled to run until the end of March 2027.

Thomas believes businesses should make use of that support while it is available.

More broadly, though, his advice to businesses still debating whether they are ready for AI is considerably simpler.

“The mistake is to wait too long,” he said.

“Treat AI as an experiment. It might take several attempts before they see a payoff, maybe, but they’ll land on, ‘actually, this thing is transformative for this business’.

“That could be in customer service, reception, marketing, or just general office productivity. There’s lots of established examples already that they can look to and try out.”

For Welsh SMEs concerned that the AI race has already passed them by, Thomas’s message is therefore less about catching up with the technology’s loudest adopters and more about finding somewhere sensible to begin.

Start small. Fix the underlying processes. Put the right safeguards in place. And only use AI where it makes the business better.

Join the conversation at Wales’ leading cross-sector cybersecurity event

CymruSec 2026 is coming to Cardiff City Stadium on 22nd October.

The Summit brings together cyber practitioners and technology leaders to examine the trends, threats, and methodology shaping the future of security.

Attendance is free for qualifying end-user delegates. But places are limited.

Register here to secure your place: CymruSec Summit 2026 | Cyber Security Conference Wales