Business and Trade Committee
Oral evidence: Artificial Intelligence, business and the future of the workforce, HC 125
Tuesday 16 June 2026
Ordered by the House of Commons to be published on 16 June 2026.
Members present: Liam Byrne (Chair); Dan Aldridge; Antonia Bance; Chris Bloore; John Cooper; Sarah Edwards; Alison Griffiths; Leigh Ingham; Justin Madders; Charlie Maynard; Mr Joshua Reynolds.
Questions 75 - 113
Witnesses
I: Karen Dewar, Chief Data and Analytics Officer, NatWest Group; Daniel Smalley, Industrial AI Lead & Business Manager for Factory Automation Digitalisation, Siemens; Kayur Rughani, Managing Director for Data and AI (UK and Ireland), Accenture; Stephen Phipson CBE, Chief Executive Officer, Make UK.
Witnesses: Karen Dewar, Daniel Smalley, Kayur Rughani and Stephen Phipson CBE.
Q75 Chair: Welcome to today’s session of the Business and Trade Committee, as we pursue our inquiry into artificial intelligence. We are looking in particular at the way artificial intelligence is or is not being diffused through the British economy today. Thank you very much indeed to our witnesses for coming along. Stephen, could you just tell us, from your perspective, how do you see AI being diffused through the businesses in the sector you represent?
Stephen Phipson: Thank you very much for the opportunity to give evidence today. We are the body representing national manufacturing. There are about 130,000 businesses in the manufacturing sector in this country, representing 2.5 million jobs, of which 800 are large companies and the rest are SMEs, so it is very important that we think about this in two parts. If we look at the data and the latest survey data about AI adoption in the manufacturing sector overall, we have 37% of that total running small pilots on AI at the moment, 43% are in that experimental stage, 18% have not done anything at all, and only 2% have full implementation of AI in factories.
Q76 Chair: How do you feel about those numbers? Do you feel comfortable that this is just an early stage?
Stephen Phipson: Let me just unpack a little further and then I can answer your question. If we think about the large companies, they are well ahead. In general there are three buckets of AI. First, generative AI is doing back office functions: helping with HR, finance, those sorts of things. Everyone is on that programme now and tend to be using that in their businesses.
Secondly, operational technology is the critical part for manufacturing, using it on the shop floor where we go more towards agentic-type AI: self-governing AI for predictive maintenance and scheduling. Here you see some very good use cases with those large companies, and we can come on to some of that data later.
Thirdly, there is compliance and regulation. We have some very good examples now where there is a lot of interest in investing to speed up the certification process for products.
Defence is probably the furthest ahead in manufacturing; in defence we have to do safety cases when we implement a new piece of equipment to the forces from the manufacturers. Typically that takes three months; using AI we now have examples of that being done in two or three days instead, so it is possible in other areas; you can imagine commercial—
Q77 Chair: Would you say that the big companies in this country are matching the global frontier? Are they basically best in class globally, or are they a touch behind or a touch ahead?
Stephen Phipson: From our recent observations on that, the Chinese are pretty well ahead with this and very well invested, to be honest with you. It is a matter of investment. We are not bad in terms of the large companies against their European competitors, largely because they have to be internationally competitive. In order for them to survive they have to be producing with the same productivity rate as their peers in other countries, so they are keeping up.
To answer your first question, Chair, the real challenge is around the SMEs. The focus needs to be on helping them adopt AI—first the digital technologies and then the AI technologies that go on top of it—and we can talk about that later with the programmes.
Q78 Chair: Let us just get a quick cross-cutting view. Karen, tell us about your group and financial services from your observations. How far advanced is the AI diffusion there?
Karen Dewar: On behalf of NatWest we believe we have a major role to play in supporting adoption for our colleagues, our customers and the wider economy. One of the ways that we seek to do this is through delivering trust in the systems that we deploy, and the capability to deploy those systems.
As an example, we have made AI tools available to all 60,000 of our colleagues, with industry-leading training uptake from those colleagues to use those tools in service of our customers. The way that we deploy it in customer experience is really driven by what our customers need. So you will see things such as us embedding AI in Cora, which is our agent in our mobile app that helps answer your everyday banking queries. We are also using AI in areas such as fraud detection. But critically, where we find AI playing a really important role is in supporting our relationship managers who serve our customers directly.
Q79 Chair: Do you think your experience is characteristic of the financial services industry?
Karen Dewar: I would like to think that we are ahead of some of our competitors, as I am sure you would expect me to say. The evidence I have for this is in the Evident AI benchmarking. We rank 16th globally, and in part that is due to the work that we have done on early experimentation, building our own infrastructure, and then giving our colleagues and our customers trust in where we are happy to use AI and where we are not.
Q80 Chair: From what I can tell you are also spending a lot of money on it.
Karen Dewar: We are investing in our own capability, absolutely.
Q81 Chair: When you get to board level conversations about the kinds of investments that you are making, what are those conversations like? Just give us a flavour.
Karen Dewar: Conversations are very rigorous. We seek to deploy AI where we believe it actually helps us differentiate our customer experience. That is either because we are able to free up the time of the people who serve customers directly, which we would refer to as supercharging our colleagues, or in our specialist areas. For example, we have around 12,000 team members who actually commit code, and 40% of our code is now generated by AI, which helps us get new propositions into the hands of our customers more quickly.
Q82 Chair: That is very interesting. Daniel, did you recognise the picture that Stephen painted in the manufacturing sector?
Daniel Smalley: Yes, absolutely. Siemens is a global technology company. We have business interests in mobility, healthcare, infrastructure, and manufacturing, so we do a lot of work in industrial business. We are a large-scale manufacturer in our own right globally, so we have experience of deployment of AI within our own manufacturing operations to make sure we can increase our productivity and reduce overall costs. These are drivers within our own operations.
We also have product businesses that support manufacturers. Exactly as Stephen was saying, the vast majority of our customer base are SMEs. We have experience with working with large multinationals, which I would say are probably further ahead, being in a position to be able to fund data science and dedicated teams focusing on these topics, but it is more of a challenge within the SME space in terms of that.
Q83 Chair: How do the conversations about AI investment go within your firm?
Daniel Smalley: Siemens is a large-scale adopter of artificial intelligence, similar to the example from Karen. We have made various models available to our own people internally, and I would say the adoption has been greatest in back-office functions: financial, legal, commercial or marketing-type functions. There has been less, or slower rates of adoption on the factory floor, largely because there are some fundamental challenges around the adoption of artificial intelligence in operational technology environments. You will hear Siemens talk quite a lot about industrial AI, trying to differentiate some of the challenges of adoption in an industrial environment from an IT environment.
Q84 Chair: Kay, just to round out the picture, you are obviously working with a lot of companies across the world. How are you finding they are taking decisions about how much to invest, whether they are investing enough, and where the returns on investment are most fruitful?
Kayur Rughani: It is a very topical question. From an Accenture perspective, we work with some of the largest enterprises, as you said. In my role, particularly focused around UK and Ireland clients, I get to see a broad range across all industry sectors, all the way from creating business cases, strategies, getting things to a stage where organisations are comfortable to commit to an investment, and then through the delivery phase. That is often where, as this technology is being built, we start to see the real challenges and how to overcome them. Where we act is really a combination of technology, the business change when it meets processes within enterprise, but also people in organisations, and how to upscale and change the roles and the models that people work in.
In terms of what we are seeing around investment, I will give you an example. There is a large financial services client that we worked with. As Karen said, it was actually a very rigorous process. It took us nine months, jointly with the client, to take a business case through approval, where the initial phase of the project was three months. There was a huge amount of focus up front on where the right area to focus was and how we could bring all the elements together, including the technology.
That is often the easy bit because you just essentially procure technology. But how do you get your business data—all the corporate information and data—ready? How do you bring your people on the journey? There is a real element of creating the right posture and the right message back into the organisation, and then getting the right business stakeholders lined up, who will quite often cut across organisational silos. When you talk about putting AI into business, you are quite often doing it in an end-to-end process that runs across different silos owned by different executives in the business, and creating that alignment.
Q85 Antonia Bance: This is a question to all the members of the panel and I will come to each of you in turn. Where have you seen the most credible productivity gains from AI so far, and are there areas where expected benefits have not materialised? I am particularly keen to understand where you are beginning to realise any savings, and where those productivity gains are real. I will start with you, Karen.
Karen Dewar: Where we see the greatest benefit from AI has been around productivity of those who serve our customers. I would like to share with you a specific example. We deployed the AI capability to automatically summarise client and customer meetings into our relationship manager teams, both in our corporate and institutional banking team and in our private banking and wealth management team. The ability to automatically summarise is actually quite well-known and trusted across AI, but there were unintended benefits from this change.
We saved 72.5% of note-taking time for a population of 220 relationship managers. The unintended benefit was that they were then able to be more present in the client conversation to identify a greater range of needs for those clients. We actually saw the client’s experience improve, as evidenced by an increase in our net promoter score system, which is what we use to measure the quality of that interaction. That is why we have such confidence that this really does supercharge our customer-facing teams.
Another benefit is that by using AI to automatically summarise those conversations we are able to see, at an organisational level, what things are most important to our clients. I can bring that to life: if you look at the emergence of conflict in Iran, for example, we can quickly understand what is concerning our clients and support our relationship managers with information that just makes the experience of the client of us as NatWest much more personalised.
Q86 Antonia Bance: You said that you had 200 or so people working in the role that you referenced. I am sure you cannot speak about individual roles, but is your intention to maintain that sort of staff profile, or to change it going forward as a result of what you have learned, and perhaps to reduce the number in that cohort of your team?
Karen Dewar: We have adopted a safe and responsible approach to rolling out some of our AI capability; you are absolutely right, we have identified a population of 220 RMs in our wealth business, and we would look to replicate this capability across the team. It is absolutely not our intention to change the make-up of that team, but rather to supercharge them to better deliver the service to our customers. It is exactly the same in our corporate and institutional space and even in retail, where we have saved almost 70,000 hours from our teams that handle incoming telephone calls, as they no longer need to capture notes from that interaction.
Daniel Smalley: I can share a couple of industrial examples from our customers; some of the stuff that is in the public domain. One is Kellanova, which is the snacks company that came out of Kellogg’s that has since been acquired by Mars. Siemens was engaged with that company around the Pringles project to optimise and reduce waste in Pringles production. People eat a lot of Pringles, so this is a multi-billion-euro business. We helped Kellanova produce something called the dough digital twin to really look at how to take variability out of the dough that goes into making Pringles. In terms of the headline figures, it involved a 10% capacity increase in the production of Pringles in the line, a 13% waste reduction, and a 7% energy efficiency gain. It really tackled the top end in terms of producing more product, but also the bottom end in terms of reducing waste in production, so it was a highly successful project.
We do a lot of work with water companies and energy companies in the UK as well. We have an example from an infrastructure perspective of an application we developed in conjunction with Yorkshire Water and the University of Sheffield, producing a solution around predicting blockages within the water network, ultimately to reduce instances of flood and sewage discharge into the network. That now allows it to predict a blockage three to four times more efficiently than it was previously able to do, ultimately allowing it to make more targeted maintenance efforts in terms of reducing that flood risk.
What is really important here is that applying this industrial AI technology is allowing us to tackle engineering and industrial problems that we had no better solution for, so we can now address new challenges.
Stephen Phipson: Building on some of Daniel’s points, from what we have seen the big prize is on that shop floor AI: the operational technology side, and fully autonomous predictive maintenance in particular. When those systems are implemented we see an increase in productivity of between 10% and 30%. Again, some of the interesting stats from our surveys show there is an overall reduction in waste of about 30%; a 50% faster time to market with new products coming out of those shop floor programmes; defect rates massively reduced; the ability to do mass customisation; and a 30% lower carbon footprint because energy efficiency is there. It is a great way of improving energy efficiency.
Just to your point there, the challenges are particularly in those plants that have older equipment and legacy data systems. This is a really big challenge for them to implement these systems, and we have quite a few of those businesses in this country. So we are seeing these kinds of results more towards the newer investments or the newer factories.
Q87 Antonia Bance: If we were talking about challenges for manufacturing more generally, Stephen, I am sure you would be telling me about energy costs, would you not?
Stephen Phipson: Absolutely No. 1; you would have seen me in the press this week doing nothing but energy. You are exactly right.
Q88 Antonia Bance: You have talked about the carbon reduction, but do you think there is real potential for AI to cut the energy needs in some of our intensive industries?
Stephen Phipson: Energy intensive is slightly different, but in general manufacturing, which is where we have a huge challenge, if we go just before the AI step and digitise manufacturing—what we call our Made Smarter programme in this country—the results of the 6,000 or so SMEs that have been through that programme show an average of about a 40% reduction in energy consumption. I was suggesting we started calling it Made Greener rather than Made Smarter because it is actually a big contributor to the problem we have on energy. That is the big issue.
Kayur Rughani: I will lean on some research Accenture has just published called Generating Impact, which is specifically aimed at the UK economy in terms of potential for AI. We surveyed over 500 executives and 2,000 employees across 17 industries, so it is a pretty rich data set backed up with some economic modelling. We are finding that only 10% of our clients or respondents are what we would call scalers. So the level of activity is huge, but actually the level of progress right at the front end, where people are scaling this and getting real benefit from it, is still relatively small. I will give you a couple of examples.
From the research, and backed up by what we see in the field as well, two of the biggest areas of benefit are software engineering, which is very native to this type of technology, and actually anything in the back office. We work with a large global immigration law firm where we have put not just AI, but actually digitisation of workflow into its process. The whole immigration and visa application process has been hugely streamlined, with AI doing things such as reading documents, checking certificates, checking backgrounds, auto-filling forms and writing cover letters that previously lawyers and other administrative staff would have done. It is one of the most successful applications at scale that we have seen.
On the other end, where things have not gone well, we are seeing that in the customer space people are being really careful. To have this type of technology operate autonomously in a customer-facing environment is quite a risky place at the moment. We see organisations trialling it, maybe in a youth brand or in a non-major brand, just to test it out and see how it goes. I certainly have an example of rolling out a very similar transcription use case in a customer-facing environment that probably did not give the benefits.
That was really a particular team that was dealing with customer duty regulations, which is a fairly regulated space. Auto-transcription was created at the end of a customer contact. The person was then actually spending twice as long comparing the auto notes with their own personal notes to create a super set of notes, and the whole exercise was taking much longer. It spiked up and showed in our performance results, and it was only when we drilled into it we realised that what really had been missed was adequate training for that team to know that they did not need to do both of those things. This is a slightly different way of capturing notes, so you really use the technology as a co-pilot rather than trying to combine both yourself.
Chair: Given these challenges, let me just bring in Dan Aldridge on the role for Government on this.
Q89 Dan Aldridge: Hi; thank you all for being here. Over the years Government and government agencies have funded lots of digital and technology adoption schemes. Thinking about outcomes of that, what has worked and what has not worked? Building on that, we know that the current Government have started to move into the AI assurance space and are really starting to take responsible AI seriously. Basically, is this a place for Government, and should they have a stronger role in setting standards and policies in this area?
Chair: Stephen, you talked about Made Smarter.
Stephen Phipson: Yes, it is probably there that I would answer the question. You are right: there are a number of programmes and three that affect the manufacturing area. First is Made Smarter, which I will talk about in a second. Secondly is really what we are doing around the Catapult system and its ports to industry, which is vitally important. Thirdly is Innovate UK, with its BridgeAI 2 programme, which is really about a front door to access the skills, training, and everything else that is there.
Going back to the 130,000 businesses, most of which are SMEs, we have to bear in mind that many of them are not even on the basic digital journey to start with, so the critical nature of Made Smarter is to get them on that journey. We have run it in nine regions around the country so far and have had about 5,000 SMEs go through the programme with really good results. The funding from Government is around £20 million this year; it goes up by another £8 million next year to address more regions. The results from that are pretty good; these are really time-constrained SMEs, maybe without the leadership knowledge about anything to do with AI or digital skills. Showing them the improvements that can be made by putting digital technology on the shop floor has been extremely beneficial.
If we want to be competitive and maintain the supply chain here that is an absolute key role for Government, but it throws up a few challenges. One is skills, which is another area that I do not think we address clearly enough yet. That is a combination of not just the data skills you need to do some of these things, which are less and less as time goes on, but also the leadership skills about being able to recognise the need for the business to do that. That is where Made Smarter has really made quite a big impact. There is a proposal called the AI advanced manufacturing adoption programme that has just been produced.
Dan Aldridge: That rolls off the tongue.
Stephen Phipson: Yes, I know, it’s a hard one. Actually I have probably said it wrong—let me correct the record. It is the AI Adoption Plan for Advanced Manufacturing, which is done by the AI champion for that part of the industrial strategy. There are eight AI champions in Government and we have worked with this one on the advanced manufacturing plan, which talks about stitching these programmes together, aligning them and trying to get better effect from them, and to get many of these SMEs on that journey. That is where we need to focus the effort.
Q90 Dan Aldridge: Is there an issue around scale and national mission on this?
Stephen Phipson: It is about doing it quicker, absolutely. It is about a national programme to do this quicker. We have to get 130 businesses through the programme, is the idea.
Dan Aldridge: Is it not 130,000?
Stephen Phipson: Yes, 130,000, sorry—you’re right.
Q91 Chair: Karen, you look poised to speak.
Karen Dewar: I was hooked by the question of whether it is a role for Government. From a NatWest perspective it is our accountability to drive responsible adoption of AI with clear guardrails and empowered colleagues. That is what drives trust for our colleagues and our customers to actually move with us on an AI adoption journey.
From a financial services perspective, specifically thinking around regulation here, we believe firmly that working through existing frameworks is a credible baseline for how we govern AI. It enables firms to innovate within clear guardrails. Priority needs to be given to clarity on application of regulation as needed, as we move forward. The one area we would call for careful thought around is whether that perimeter is in the right place. I am thinking here about the fact that we are likely to see increased use of AI tools for financial advice. We need to ensure that the perimeter is thoughtfully positioned in the right place to protect all our clients and customers who may, in fact, go to that tool for their advice.
Q92 Dan Aldridge: Can we build on that slightly with the assurance piece? Does NatWest have an assurance process around utilising AI, and is that going to build into what the UK Government do around assurance?
Karen Dewar: Absolutely. We have a deliberate, responsible approach to AI adoption. It is key to managing risks. It gives us absolute certainty on better customer outcomes and more successful adoption, ultimately. We think about it in three stages, quite simply: first, should I do this? Which helps us address the ethics of the solution that we want to put in place, and we use our own clear AI codes and an AI and data ethics panel. Secondly, can I do this? That helps us very clearly evidence that the system or the solution works mathematically. Thirdly, is it still working? That is where we are able to deploy guardrails that are data-led, specific to the use case that give us the absolute comfort and control that it continues to work in the way that it is designed. We have been working with the FCA in areas such as live testing, as you know, and we would like to see that partnership continue.
Chair: That gives us a good set of factors to help us think through the role of Government.
Q93 Justin Madders: My first question is to you, Stephen, just thinking about the next three to five years. Obviously you have a very broad range of members in your organisation; where do you see the greatest potential for improvements using AI?
Stephen Phipson: From what we have seen so far the greatest potential is on the shop floor, actually in the operational side of the businesses, and it is helping those SMEs on that journey. That is really where we have an opportunity. On some of the projects we use now as case studies or use cases with the larger manufacturers that have implemented this, we actually see employment increase as a result, not decrease, because they are more productive. That means you need more support services, more people loading materials, and more of other things. When there is an average 20% improvement in productivity they are dealing with much bigger volumes. Their overall business is much more productive and profitable, and in general what is happening is they are actually increasing employment on the shop floor as a result.
Q94 Justin Madders: Do you think that if you replicate what you have already done on the big ones, it will translate into smaller businesses?
Stephen Phipson: I absolutely believe that. We have seen some early results with Made Smarter, but we need to go a lot quicker with helping the SMEs get on, because that is where the real challenge and the opportunity is; that is where most of the employment is.
Q95 Justin Madders: You have already mentioned some challenges in delivering that, but are there any others that you have not yet mentioned?
Stephen Phipson: There are two big issues there. The first is legacy systems and how you upgrade old equipment on the shop floor and add sensors and all the other things to it; the second is about leadership, training and skills.
Q96 Justin Madders: Just so we are clear, when you are talking about legacy equipment, is there a date before which this stuff just does not work?
Stephen Phipson: I wish it was as clear cut as that, but of course it is not and it depends on where you are. You can imagine there are some sectors of manufacturing that have pretty old equipment, and we actually need to modernise what they are using. But there are solutions for that; the Catapults are particularly good at helping people with legacy equipment adopt these digital technologies. That is why aligning what we are doing with the Catapults, Innovate UK and Made Smarter together would be a very powerful statement by Government of support for the sector.
Q97 Justin Madders: I put the same question to Karen. From what you have said, it sounds like you have made massive gains already. Is there much left to do?
Karen Dewar: The areas of greatest potential really fall into those two core themes: customer focus and improving that customer experience and, as I think Kay touched on, our ability to be able to get new propositions into the hands of customers more quickly by the improvements that we see in our ability to deploy new digital propositions through the improvement in software engineering.
Q98 Justin Madders: You have obviously thought about this across the entire business. Is it about the software not being quite where it needs to be? Is it about just trying to ask the right questions? Is it just a question of time, or is there any particular reason why you will not achieve everything that you have set out at this stage?
Karen Dewar: If I have understood your question correctly, you are asking me what the barriers or the potential impediments are that I could see. I am thoughtful around three areas. There is a really rapid pace of change; the model choice that we have improves every month, and that is one of the reasons that we have built some of our own internal infrastructure.
We see some regulatory complexities—the EU AI Act versus the UK approach—and, referring back to Daniel’s comments, we see the skills gap as a potential future barrier. In the FS Skills Commission report we noted a 35% skills gap between demand and supply. Those are the three areas we consider to be barriers in this achieving its full potential for our customers.
Daniel Smalley: We certainly see some cultural challenges around risk appetites and things like that but, as a general comment, engineers are good at solving problems when given the right environments to do so. You will not be surprised to hear me talk about energy costs; this really is a topic that comes up again and again within our customer base.
I would probably expand skills and workforce capability out into really supported ecosystems. The UK is quite good at attracting AI talent in certain sectors, largely because we create that stickiness so that there are more jobs in certain sectors and more universities and institutions supporting these, rather than it being an isolated thing. That does not really exist within manufacturing, so it needs to be nurtured and encouraged. Of course, access to finance to adopt digital technologies is not an AI-specific question or a new challenge for manufacturing. We have probably had the same conversations around automation and robotics and these things in the past. The technology is there; it is how we scale adoption of it.
Q99 Sarah Edwards: Daniel, you might be able to provide a bit of a wider context. Obviously Siemens is global and operates worldwide; is there any light that you can shed on how AI adoption is happening across the Siemens group, and is it different in different countries? Has it taken a different approach, and are there any things that you can cite, or particular jurisdictions where you would say, “This is actually better or working really well,” or any pieces of advice where you would say, “This is actually quite difficult, and therefore we have perhaps struggled here.”? As you are working across so many different areas, are you able to give us a bit of insight into how it works?
Daniel Smalley: We would like to consider ourselves a leader in this industrial AI space. It is a topic Siemens has been involved in, from an R&D perspective, since the 1970s. It is only probably since 2022, with the advent of ChatGPT and low-cost compute, that we really started accelerating deployment of some of this stuff into the shop floor.
In terms of the global picture, the US and China are clearly leading in that adoption for a number of reasons in terms of investments and really a long-term strategy supporting this.
Q100 Sarah Edwards: This is within your Siemens group, so are those the two jurisdictions where you are ploughing ahead much faster internally? I am not talking about their Government policies; I am more talking about how you approach it as a company.
Daniel Smalley: That would be within our customer base. Within our own group we are, of course, under control to be able to accelerate adoption within our own factories. You would certainly find a lot of AI deployment use cases within our own manufacturing facilities in Germany, but in the UK as well. We manufacture electronics up in Congleton, and you will find use cases around AI adoption there because they are constantly trying to remain productive on a global stage, competing against factories across the Siemens networks. This constant investment in the people and technologies to retain that competitive advantage is how we, as a higher-cost country, can remain competitive. So yes, this is a constant.
Q101 Sarah Edwards: Are you using different pieces of AI in different places, here and in Germany, or are you using the same tools globally? How do you approach it: in a very niche way, or a blanket roll-out?
Daniel Smalley: AI is a very broad umbrella term that Siemens is involved in across the board, but that scalability and focus on infrastructure is key to how we are scaling out the same use cases in other factories.
As with a lot of our customers, there is a lot of piloting and a lot of testing going on. What is often missed in the piloting is how you get out of a pilot and deploy 10 times, 100 times, or 1,000 times. That is one of the key learnings in terms of adoptions within our own factories: that we have to have common infrastructure so that we can really say, “We can solve a problem in Germany, and we can roll it out in 100 factories elsewhere.”
Q102 Chair: To follow-up on Mr Madder’s question, he was basically asking about the kind of frictions that are slowing adoption. When you look across your implementations, are you spotting things that are in common?
Kayur Rughani: Absolutely. Nobody is short of ideas; there are plenty of ideas in industry where AI can apply. The biggest barriers are legacy technology, which was talked about in terms of the manufacturing context, but it exists even in professional services organisations or large enterprise. These AI systems, at some point, have to interface with the machines and the infrastructure that already exist, and in some places that is just not up to scale; it will not be able to cope, or it does not have the interfaces available.
Secondly is the data because otherwise the technology is very general purpose. You can deploy in five different organisations, but the only way you are going to get any form of competitive advantage or, probably more importantly, accuracy, is by teaching it from the data that you have. That is a real effort in many organisations, as data is stored all over the place—in documents and in various people’s folders—and you need to bring that together to govern and manage it, and bring that in.
Thirdly is really just around the skills piece. The survey that we did was that 18% of employees that we surveyed use generative AI tools every single day at work—triple what it was just 18 months ago—but 62% of those workers have had no training from their employers. So a big population is just figuring it out themselves, where actually a real focus on very role-specific training would make a huge difference.
Q103 Alison Griffiths: I was quite interested to hear, in your individual companies and organisations, and among your members, whether companies are coalescing around a smaller number of providers or whether there is a genuine plethora of AI providers being used? Is there an inherent risk in having a small number of providers, or whether we are mitigating that? I will start with Kay.
Kayur Rughani: AI is absolutely everywhere in the clients that we work with; every single piece of software that they already have in their enterprise now has an AI bolt-on, and they are expected to pay licence fees, so it suddenly becomes a very important topic to discuss.
The way the AI technology is changing, given we are still at a relatively early stage in the maturity of it, it is wise to have multiple options. We would certainly be nervous of advising a client if they are narrowing down too quickly. The tool of the day at the moment is Anthropic’s Claude tool, but in six months’ time it could be something else. The way the technology is moving, that can happen.
Stephen Phipson: A good way of characterising it is if you think about three areas. So in the back office functions, HR and accounting, there is a plethora of different tools to be used. There is no particular dominant technology that is used in the general AI space, and that changes all the time, as Kay said.
It is a lot more limited on the shop floor, where Siemens is a major player. It tends to be the integrators that provide the automation for the shop floor that are then moving into the AI space and helping with that. There are not many of those companies; Siemens and a few others like that are very typical of that space.
Some very large companies are writing their own programmes and AI solutions from a corporate perspective and then deploying it around the world. Car companies are tending to do that: in a corporate centre they will have a generic, “This is the way that we do this,” and then they will deploy it in 50 factories around the world afterwards. But really the biggest opportunities for the SMEs are in the centre, in that OT opportunity, and there are a couple of really big providers.
Q104 Alison Griffiths: We have a session on SMEs after this, so I want to make sure we focus on major sectors, which is what the focus of this session is. But thank you for those answers; that was actually really helpful for me.
The main thing I actually wanted to ask you about, and some of you have already mentioned it, are the factors driving successful AI adoption by sector. We have talked generally about leadership, skills and other factors, but thinking about the sectors that are being more or less successful in delivering AI and why, what are the reasons for different firms being able to move faster or slower? I will start with you, Daniel.
Daniel Smalley: There are differences in terms of levels of adoption because there are differences in levels of adoption of digital and automation technology in certain sectors. You may find that automotive, aerospace, and pharmaceutical industries generally have been adopting digital technologies for a long time, so it is a lot easier for them to deploy artificial intelligence over that infrastructure.
Q105 Alison Griffiths: Picking up on the point that Kay made earlier, would they be more likely to not have legacy infrastructure, and to have upgraded their infrastructure already?
Daniel Smalley: Yes, but arguably the larger sectors, such as food and beverages, is where there is a lot of value to be gained out of it. These may be sectors that have not invested, but there is an opportunity then for step change and to leap ahead in terms of making that investment in digital technologies so that they can then gather data and apply artificial intelligence on top of that.
Q106 Alison Griffiths: Karen, would you like to speak to your customer base? I am conscious you are in one particular sector, which is financial services; otherwise I could ask the question to Stephen and Kay if you prefer?
Karen Dewar: I would probably be more comfortable to answer for financial services. I cannot stress enough that responsible adoption helps us to manage risks and leads to better customer outcomes, and that establishes trust. We actually need our customers and our colleagues to trust the solution to drive that adoption and see the real benefits. Again, I would say having clear guardrails and frameworks manages those risks. We certainly have central capabilities, but visible top-down sponsorship and local empowerment, and those would be the real lessons I would share across financial services.
Chair: We must be brief because we have a vote coming, and there is one more question I am going to try to get in before the bell goes. But Stephen, please proceed.
Stephen Phipson: I will be quick then. One would be the influence of prime contractors, the big organisations, on their supply chain is an important factor. So in the car company sort of area, the automotive sector, we have thousands of SMEs in those supply chains. The influence of those large companies, not to force them but to encourage them to adopt, or to insist that they are in the process of being more competitive, is leading to adoption there. The other one is Made Smarter. I cannot explain enough just how much of an effect that programme has had. That is one of the really positive things, and we just need to do much more of it.
Q107 Alison Griffiths: Kay, do you also feel that adoption rates between businesses differ between those that deal directly with customers and those between digital native firms and the more traditional ones, in your experience?
Kayur Rughani: Yes absolutely—certainly because a digital native firm will have the inherent muscle inside the organisation to be able to adopt this technology. But what we are seeing from our research is banking insurance and software-type companies are by far the biggest adopters at the moment. Part of that is also they are used to working in this regulated space, and this type of technology lends itself to being able to operate in that.
But one of the most important factors is knowing your process—having people in your teams who can rethink a way a particular business process will work and then putting that technology not necessarily in old processes, but in a new, re-imagined process.
Q108 Leigh Ingham: I am going to be as quick as I can. One of the things that is really important to understand is how AI is changing tasks, skills requirements and ways of working within your business. I know we have a really short time, but could you make sure you home in on young people and early career entrants within your answer, as that is something we are particularly interested in, or certainly I am.
Stephen Phipson: Digital skills are really important. We are finding that younger people already have some of the basic skills to deal with this and actually a lot of manufacturing businesses are interested in employing them to help them on that journey. Traditionally engineers are not an old grizzly bunch, but they are pretty conservative bunch of engineering-type characters, as we know from the sector. They actually like the fact that someone is thinking differently, and a lot of the younger people joining these organisations have those digital skills.
The other challenge is leadership skills. It is about recognition that we need to change and adopt a different business process or culture in the business. A lot of these are small businesses that have been in business a long time: first, second, sometimes third generation-type manufacturing businesses. That is where you need the external assistance from something like the Made Smarter programme.
Kayur Rughani: I run the data and AI business within my department in Accenture. We have a centre for advanced AI, which we are trying to grow. We have seen graduates and junior talent coming in through all sorts of schemes, and within 18 months they are able to leapfrog people in front of them. So we have had a really positive experience of people who are very comfortable with this technology.
We are also thinking first principles; going back to my point about designing process, there is no point in just putting a technology into a process that has been the same for 25, 30 years. You have an opportunity to redesign something and be able to think on first principles basis, so we are going to be increasing the number of graduates that we take on, year on year.
Karen Dewar: I would say technology re-shaping work is not new. I have worked in NatWest for 28 years, and when I first joined we still used a fax machine. Indeed, it was only 2010 when we launched our first fully functioning banking app.
It is our responsibility to equip all our colleagues with the right skills for the future. I am really keen to reflect Kay’s point on that positive impact that we see from young people, and for that reason we are investing heavily in early careers. We had the largest graduate cohort in 2025, and again we are hiring 1,300 grads and apprentices this year because we can really see that positive impact.
Daniel Smalley: From a Siemens perspective, similarly we are not having to invest a lot of time in terms of teaching our early-career professionals about the benefits of digital technology and artificial intelligence. These are really the sorts of tools that they start with natively. They are not trying to apply new technology to old processes; they are really looking at things differently. I would say there is quite a lot of opportunity in terms of focusing on young people within the business to help accelerate adoption.
Leigh Ingham: Do I have time for a follow-up, Chair?
Q109 Chair: The vote is not too far away, but I just want to conclude with just a final reflection on Dan Aldridge’s question, if you could give just a very quick-fire answer to this. We have talked about things that Government can do, and we have highlighted issues around legacy, data, skills, and energy costs. Stephen, you have underlined the virtues of Make Smarter. If there is anything else that you would put on the list of helpful Government interventions—not all Government interventions are helpful—what would you add?
Kayur Rughani: The only other one I would call out, and this is more for Government adopting this type of technology, is procurement. We still see procurement processes that are not set up to bring out the benefits of this technology.
Q110 Chair: Yes, okay. Karen, you have mentioned regulation and clarifying that regulatory perimeter, do you want to add anything else?
Karen Dewar: I would say continuing to work together on the skills gap, but also where we have had real success and leveraged our position as the largest bank for business is really engaging with the SME population that both Stephen and Daniel have spoken about, and educating and trying to bring it to life in person across our 14 accelerator hubs. So I would add really helping to make practical advice and support available to all the SMEs as we go through this AI adoption journey.
Q111 Chair: Are there other things that Government can do to jump-start SMEs, for example around the tax system or working through the accountancy profession? Are there any other kind of contact points that SMEs have to have, that could be used as opportunities?
Karen Dewar: Absolutely. One of the things that we did was a strategic partnership across ourselves and Google Cloud. That means that we are using our experts and Google experts to go out and really tangibly demonstrate how you can deploy AI to make your business more healthy and constructive. We should seek to leverage any interaction points as a combined team.
Q112 Chair: Okay. Daniel, apart from collapsing energy costs, do you have anything to add?
Daniel Smalley: Yes, we are supportive of the points made in the AI adoption plan that Stephen referenced from the High Value Manufacturing Catapults authored By Professor Dungey that makes a lot of points in terms of how we adopt. We clearly see this is not an innovation problem around new technology; it is the adoption of technology.
Q113 Chair: Okay. Stephen, have we missed anything from the list?
Stephen Phipson: The same point as Daniel really: it is that alignment, that adoption programme. We are very supportive of what Chris has come up with and we think that that programme is the way forward to try to amplify and align—and, of course, then reduce energy costs.
Chair: Reduce energy costs, yes—we have that point. Great. That has been a brilliant discussion. Thank you very much for getting us started.