The case for SME adoption is well rehearsed. SMEs account for around 62.3% of employment in Wales, adoption of AI among them remains low, with estimates around 7 to 10% and Senedd-commissioned research finding 7.4% in 2022, and Wales has the lowest productivity of any UK nation or region. If AI does not reach the firms where most people work, it will not change Wales's economic trajectory. All of that is now widely accepted.

The harder question is the design one. What does a serious adoption model actually consist of? Grants and awareness campaigns are not enough, and short pilots tend to generate activity rather than change. A practical model has four parts that fit together: a way in, a way to pay for it, a way to deliver it, and a way to know whether it worked.

From funding to firms

The first design choice is how public money reaches businesses. Each Growth Zone comes with adoption and skills funding, and the temptation is to spend it on broad, easy-to-announce activity. The framework's second test argues for the opposite: structure funding around real SME projects, not pilots. That means backing focused transformation projects in live businesses, organised so that the money follows a defined problem to a measured result.

A good model also meets firms where they are. It should have a simple front door that routes a business from an initial enquiry into a funded project within weeks, not months, with light-touch paperwork and pre-approved providers. Complexity is itself a barrier: many SMEs simply do not have the spare capacity to navigate a bureaucratic scheme, however well funded.

Cohorts and vouchers done well

Sector-based cohorts are one of the most powerful tools available. Instead of treating each business as an isolated case, a cohort brings similar firms, in manufacturing, food and drink, care, tourism, logistics or construction, through a structured process together. They compare use cases, share trusted providers, and reduce the perceived risk of adoption. Cohorts also give delivery partners and government a much clearer picture of what is genuinely working in a sector.

But cohorts must be more than networking groups. They should be built around delivery milestones, so each firm leaves with a defined use case, an implementation plan, and a way to measure impact. Alongside them, voucher-style funding can let SMEs buy practical help from accredited Welsh providers rather than being forced through a single central programme. Done well, vouchers both support adoption and grow the Welsh AI services ecosystem. The essential condition is that they are tied to outcomes, not just spend: the test is whether the support helped a firm make a measurable change.

Hands-on delivery, not just advice

The part most adoption programmes underestimate is delivery. Assessments, training catalogues and introductions are useful, but there is a chasm between a readiness report and a working AI capability embedded in a firm's daily operations. Crossing that chasm takes people who will do the work inside the business: understand the operating model, fix the data foundation, build and integrate the solution, train the staff, and stay long enough to see whether it holds.

This is the case for a forward-deployed delivery capability sitting behind the front door, explored in more detail on the Delivery page. Small multidisciplinary teams work directly inside Welsh SMEs, delivering exactly one prioritised, high-value intervention per firm through a consistent journey, with a baseline taken before and the same indicators tracked after. Crucially, delivery should run through an accredited network of Welsh providers working to one shared playbook, so capability and cost improve over time and the money helps build local supply rather than importing a delivery machine that leaves when the contract ends.

Two things must be designed in from the start. The first is a rural and Welsh-language lens, so support reaches firms outside the main commercial centres and works in the language businesses actually operate in. The second is a clear exit: firms should be able to operate the capability themselves, take light-touch support, or buy a managed service, with no forced dependency afterwards.

Measuring what changed

The final part is measurement, and it is what separates a serious model from a well-meaning one. Every engagement should establish a baseline before implementation and track the same indicators afterwards. The right question is not "did the firm use AI?" but "what changed in the business because it did?"

Those measures are practical and familiar to any business owner: did process time fall, did output per employee rise, did quality improve, did errors reduce, did margins increase, did staff move into higher-value work, did the firm win work it could not win before? Alongside firm-level results, the programme should track its own health, rural participation, the share of spend reaching Welsh suppliers, and the reusable templates it creates, and it should publish aggregate outcomes openly.

Put together, these four parts, a simple way in, outcome-tied funding, hands-on delivery, and honest measurement, describe an adoption model that could actually move the numbers. It is more demanding than running a series of pilots, but it is the difference between activity and transformation, and it is the version of SME adoption worth funding.