Our new Director of Policy and Skills, Professor Ashmita Randhawa's first blog on AI skills, development and training
Solving problems first, developing skills second
Most organisations that come to the UK’s National Innovation Centre for Data (NICD) do not arrive asking for training. They arrive with a problem: a forecasting model that keeps drifting, a maintenance regime built on guesswork, a brilliant idea for a new AI-driven product or service that they lack the capability to design or build, a service where the data exists but nobody can answer the question being asked of it. Training and skills development is what happens on the way to solving those business problems.
That ordering is not an operational detail. It is the whole argument about AI skills, compressed into how our centre chooses to work, and it runs in contrast to how people normally think about the ways in which to develop skills, particularly in the way that policies on skills development are designed.
The limits of a supply-side approach to skills
Leading academics (see the work of the SKOPE research centre) and practitioners (Colleges and private training providers) across the skills landscape have long (and I mean very long) argued that skills development cannot be thought of as just a supply side problem. For too long policies or interventions have invested heavily in producing skills without the recognition of employer demand or indeed about the utilisation of the skills that are produced by those supply side interventions.
As Ewart Keep puts it, a supply only model of thinking, without thinking about the whole system within which a person traverses the world of work addresses on a third of the problem, leaving demand and utilisation of those skills entirely untouched – a half (or a third, if you will) attempt at solving skills crises.
Apply that thinking to the current narrative on AI skills; we once again find ourselves facing a very high risk driven by the supply side thinking described above. We could fund the development of AI skills at the national level and run the risk of seeing very little return, if we don’t think about what it is that firms actually want in their hires or even pause to consider the binding constraints that often sit in firms. Businesses, large and small that may not have changed a single process, job profile or management proactive to make sure those skills can actually be utilised.
Image of Senior Data Scientist Louise Braithwaite presenting slides to businesses at an event.
Turning demand into skills development
The NICD model is interesting precisely because the thinking on skills development has been built the other way around.
Take the example of the NICD skills transfer projects - a collaborative team of NICD data scientists and the client's own staff works on the client's real business problem, on their data, on their infrastructure, over a sustained period of time. Skills transfer into the organisation happens organically as a by-product of solving something the organisation already cared about enough to commit staff time to.
Demand for those skills is not forecast; it is the entry condition required to solve the problem. Utilisation is not hoped for; it is the medium the learning happens in.
Why AI literacy must be contextual
In 2025, the Skills England report on AI skills for the UK workforce showed that most workers in the UK economy will require AI literacy (i.e. the ability to use, verify, and safely integrate AI tools), whilst a smaller share will need specialist technical skills.
One cannot teach verification, judgement, and responsible use in the abstract. They are situated capabilities: they mean something in relation to a specific dataset, a specific decision, a specific set of consequences if you get it wrong.
A team that has spent months arguing about whether their own model's output is trustworthy has learned something that a short intervention cannot easily replicate; or at least an intervention that does provide that very opportunity to students.
Learning through real projects: The NICD AI traineeship
This is exactly what the NICD AI traineeship model has tried to do.
Funded by DSIT in 2025, the six-month programme combined technical and professional development with hands-on experience of delivering real projects. Rather than working on theory and simulated ‘real-world like’ problems, the trainees, hand in hand with experienced data scientists at NICD worked on real business problems put forward by SMEs in the region.
By proposing these projects, the SMEs were no longer acting as passive customers of an education system; they were behaving like active partners inside it – something that has long been called for.
From the trials of delivery to project management, to learning how to balance the demands of a client to the realities of how data can actually be wrangled; the trainees gained the employability skills and ways of working that employers really need of their data scientists. Suffice to say that over 70% of the trainees are now in good jobs, applying these skills – no small feat.
Students carrying out field work as part of their traineeship with NICD.
AI adoption and AI skills: Two sides of the same coin
Whilst the models described above do indeed start to address some of the missing gaps in skills development, it is only honest to say that these models are intensive, and scaling such models is not easy, and requires an appetite for risk to address the skills problems we face; indeed, thinking about a whole systems approach.
What the model does demonstrate is that there is a need to ensure that we are continuing to fund the demand side as a skills intervention. I have long argued that digital (in this case AI) adoption and skills development are like two sides of the same coin.
Adoption support, management capability and work redesign are not adjacent to AI skills policy. They are the conditions under which AI skills have any value at all. So, if we are going to be serious about supporting the development of AI skills, we have to be serious about how we support firms to adopt AI in the ways that are meaningful to them.
This means that national organisations such as NICD need mandates and multi-year funding, and not project by project survival, to ensure that we are able to develop the AI skills that are truly needed.
Demonstrating impact and building the case for scale
Scale can only be justified when there is demonstrable impact of interventions and that is exactly what NICD can do.
Based on an independent economic evaluation of the projects that NICD has run over the last 9 years, we are able to demonstrate excellent value for money to the public purse, generating £27.20 of net additional GVA for every £1 of direct public funding” [≈ four times UKRI average]; not to mention the number of jobs created and the number of people upskilled.
But if I share more, I will spoil the reveal that is the ORTUS report, which will be published next week at our event on Wednesday 16th September. Do join us where we can continue our debate and discourse, and learn more about the impact of our national innovation centre.
You can read more of our news stories and sign up to our newsletter to keep up to date with our latest news, events and developments.
