Subscribe to our newsletter to receive business tips, learn about new funding programmes, join upcoming events, take e-learning courses, and more.
Take the first step to successfully starting and growing your business.
Join for free)
Posted: Tue 22nd Sep 2026
6 min read
Last week I spent the day at the Government Technology and AI Conference 2026, chaired by Professor Matthew Forshaw, Senior Advisor at the Alan Turing Institute and Professor of Artificial Intelligence at Newcastle University.
Matthew set the day up with four questions, and I'd suggest any small business owner thinking about AI writes them on a Post-it:
What would meaningful improvement look like, and how would we know?
What do our people need to be able to do?
What evidence would give us confidence that a system is fit for its intended use?
And who benefits?
His framing line is the one I keep coming back to: AI is one tool among many, and sometimes better data quality or a simpler process delivers more value.
That's not a fashionable thing to say in 2026, but it's true, and it will save a lot of businesses a lot of money.
Professor Nisreen Ameen, AI policy adviser and director, Digital Organisation and Society (DOS) Research Centre, Royal Holloway, University of London, shared findings from two reports published with Skills England, built on 23 expert workshops, 150 organisations and over 500 employer survey responses.
Two numbers stood out. Two thirds of workers already use AI that's built into tools they share with colleagues. Forty-four per cent of organisations report daily AI use.
And yet capability isn't keeping pace. Training, where it exists at all, is informal and unstructured, with people figuring it out between tasks.
Nisreen identified the barriers clearly: inconsistent language around what "AI skills" even means; employers focusing on tools rather than job-relevant applications; fragmented provision and outdated curricula.
What I liked was that she didn't stop at the diagnosis. Her team has built three things for employers, SMEs specifically in mind:
an AI Skills Framework mapping skills by job level
an AI Skills Adoption Pathway running from awareness through to scaling
an Employer AI Adoption Checklist you can self-assess against
The training principles she set out (practical, reachable, integrated, modular, expandable, sustainable) are a fair test to apply to anything you're being sold.
The detail that got the most nods in the room was "reverse mentoring" – senior leaders meeting monthly with junior staff who already know the tools well.
Cheap, fast, and it fixes the two problems at once.
Eamonn Hennessy from the London Borough of Waltham Forest gave a fantastic session, covering a case study that was both relevant and helpful.
Sixteen automated processes across revenues and benefits, 51,000 transactions in six months; around 17,000 officer hours saved made for compelling stats.
But his rule for deciding what to automate is what small businesses should steal.
Three questions before you touch a process:
Is the input reasonably consistent?
Can the rules be written down?
Do we know what happens when a case falls outside of those rules? If the answer to the third is no, you're not ready.
He was equally blunt about the numbers. Saved capacity is not the same as saved cash. It only becomes a budget saving when you make an actual staffing decision.
As he put it:
"Credible benefits reporting is more persuasive than claiming the largest possible number."
His biggest saving, incidentally, £2.3m on council tax support, came from evidence-led policy redesign, not from automation at all. Proving Matthew's opening point.
Dr Asieh Hosseini Tabaghdehi of Brunel University has spent five years working with SMEs through the Help to Grow programme, and her message was simple:
"Join your local ecosystem (your chamber, your university) rather than working in a silo."
Her five pillars of SME concern were strategy and leadership, skills, trust, data governance, and accountability. Note how few of those are technical.
Elizabeth Anderson from the Digital Poverty Alliance brought the reality check. One in seven UK adults lives in digital poverty. One in five can't use government services online without support.
If you're designing a customer journey, her advice applies directly: design for low-spec devices, use plain language and test with genuinely low-confidence users, not your most capable customers.
Haseeb Khan from Granicus closed with the distinction I'll be using from now on: being AI-able (you have the tools, you can run a pilot) versus AI-ready (clear ownership, current data, agreed boundaries, defined success criteria).
I liked his measurement rule:
"Agree the metric before you launch, not after. High interaction volume doesn't mean a better service."
So, before your next AI decision, answer Matthew's first question:
"What would meaningful improvement look like here and how would you know if you got it?"
