How to turn AI pilots into real business value

Summary

The Times Tech Podcast
Episode: How to turn AI pilots into real business value
Date: June 11, 2026
Host: Katie Prescott (The Sunday Times)
Guests: Paul Henninger (UK Head of Technology and Data, Global AI Leader, KPMG) & Professor Alan Brown (University of Exeter Business School, Author of “Making AI Work for Britain”)

---

Episode Overview

This episode explores the challenges and realities of converting experimental AI pilots into genuine business transformation and economic value. Host Katie Prescott is joined by AI and business experts Paul Henninger and Professor Alan Brown, who share on-the-ground insights about AI’s impact in organizations, what’s stopping pilot projects from scaling, and where business leaders should focus their efforts as AI accelerates into the mainstream.

---

Key Discussion Points and Insights

1. The Hype Versus Reality of AI in Business

• Gap Between Investment and Real Change
  Many organizations are excited about AI, investing heavily, but find translating pilots into real operational and financial impact to be very challenging.

  • Katie Prescott: “Is AI having an effect yet? Is it making businesses do things better and does it match up to the hype?” [00:04]

• Integrating AI with Day-to-Day Business
  Alan Brown argues that true value only comes when innovation is tied directly to daily objectives and business needs.

  • Alan Brown: “We have this kind of idea that there's a small group of privileged people with the beanbags and the slides... and there's everybody else keeping the lights on. And I think we have to start with the point of view of no, everybody is changing all at the same time, perhaps at different rates… But everybody is part of that change.” [01:48]

2. From “Tech Change” to “Business Change”

• AI Alone Won’t Transform the Business Paul Henninger emphasizes that technical automation is not equivalent to business innovation.

  • Paul Henninger: “You can have… AI doing things like automating 80% of financial planning… But if you don’t adapt the business, the processes, the people, the teams, the governance around it, you haven’t actually changed how the business functions." [03:19]
  • There's a further step needed: converting increased capacity into outcomes that impact the bottom line, such as increased revenue or reduced costs.

• Examples: Risk Management and Mortgages
  Henninger cites risk management and mortgage underwriting as areas where AI is creating capacity, not always direct value.

  • In risk, automated controls now respond to legislation, making departments 50% more efficient, but whether that generates direct financial value depends on leadership choices.
  • In mortgage underwriting, a financial institution used AI to qualify more customers (choose growth), but the value only came after setting a clear business objective. [08:38]

3. The “Value Fallacy” and the Complexity of Real-World Impact

• Creating Value vs Realizing Value

  • Alan Brown: “This idea of creating value and then realizing that value are two different things... Where does that [freed-up time] get deployed and how does that get recognized is a different problem.” [06:53]
  • Governments and businesses often over-simplify projected savings (e.g., automating admin work), but translation to the bottom line is complex.

• Systemic Bottlenecks and Weakest Links

  • Technology often improves one area, but broader organizational systems and processes may be bottlenecks or even resist change (e.g., healthcare diagnostics vs scheduling and paperwork). [09:45-11:46]
    • Brown: “Often these tasks that we can automate... are linked to many other things... You know, I went for a scan recently... most extraordinary modern technologies... and then it takes three weeks for somebody to look at it and they lost the paperwork... All of these systems around us become part of the challenge...” [09:45]

4. Why Most AI Pilots Fail to Scale

• Not a Technology Issue Alone
  Henninger points out that most failures are due to data, people, finance, and unclear governance—not because the technology doesn’t work.

  • Paul Henninger: “I’ve seen pilots that don’t get into production simply because it was unclear who could authorize that.” [12:53]

• Success Rate Expectations
  Even a 5% success rate on pilots is robust in tech innovation. Sometimes, highly successful pilots are paradoxically the hardest to operationalize due to the radical change they require.

• Software as a Case Study
  Brown notes that while AI can generate lots of code, most of it never reaches deployment because organizations can't consume change at that speed.

  • Alan Brown: “You can’t change things 10 times a day even if you can generate code 10 times a day.” [15:05]

5. Perspective on the AI “Bubble” and Historical Parallels

• AI: Bubble or Next Industrial Revolution?

  • Both guests see analogies to the Internet boom: overinvestment, initial hype, and a later phase where foundational infrastructure and new ways of working remain.
    • Alan Brown: “Is AI a bubble? My answer is, of course it’s a bubble... What’s interesting is... what will happen when the bubble bursts? What will be left behind and what will that mean for us?” [17:36]
    • Paul Henninger: “We're doing things like buying steam engines and rocking up to a finance organization and saying, where do I plug it in?... There's a massive amount of work to do.” [17:06]

• “The Future is Not Evenly Distributed”

  • Both note that, like the early Internet era, some sectors/regions are leaping ahead with AI while others lag—often due to regulator, sectoral, or cultural inertia.
    • Paul Henninger: [William Gibson quote] “The future is already here. It's just not evenly distributed. And that's I think what we're seeing and I think the rate of change is unbelievably high.” [20:22]

6. Practical Advice for Business Leaders

• Holistic Readiness: Five Pillars

  • Infrastructure, data, people, financial measurement, and governance—all need reworking to sustain AI transformation.
  • Henninger: “The way this technology works, even just mathematically, is by setting an objective... Until you decide how you want your business to change, the AI... doesn’t know what to do.” [21:31]

• Objectives Must Be Clear

  • Leaders (not the AI) must set precise objectives, make trade-offs, and be willing to address uncomfortable choices between people, growth, profit, etc.

---

Notable Quotes and Memorable Moments

• “Often we separate the day to day business from changing the business. And we have this kind of idea that there's a small group of privileged people with the beanbags and the slides... No, everybody is changing all at the same time…”
  — Professor Alan Brown [01:48]

• “There's a difference between technology change and business change... Even when AI works, the business doesn't natively change.”
  — Paul Henninger [03:19]

• “You can have a 10x improvement in this one area, yes, but everything else doesn't have that ability to flex.”
  — Alan Brown [09:45]

• “The more successful the technology is, the harder it is to move it out of a pilot mode.”
  — Paul Henninger [13:49]

• "We always overplay. But what's interesting is to ask yourself, what will happen when the bubble bursts? What will be left behind and what will that mean for us?"
  — Alan Brown [17:36]

• “The future is already here. It's just not evenly distributed.”
  — Paul Henninger [20:22]

---

Noteworthy Timestamps

• [01:48] — The “privileged few” and the integration myth
• [03:19] — Tech change vs true business change
• [06:53] — The value fallacy: creating vs realizing value
• [09:45] — “Weakest link” and adjacent bottlenecks in healthcare
• [12:53] — Why do AI pilots fail? (governance, organization, not tech)
• [15:05] — AI in software: code generation outpaces organizational change
• [17:36] — Is AI a bubble? Parallels with the early Internet
• [20:22] — “The future is not evenly distributed”: where AI is already mainstream
• [21:31] — Practical advice: start with clear, detailed objectives

---

Summary Takeaways

• AI’s promise in business is real but uneven and highly dependent on aligning business objectives, people, systems, and processes—not just installing new tech.
• Successful AI implementation requires holistic organizational redesign—not isolated pilots.
• Hype and overinvestment (the “AI bubble”) are typical, but foundational change will endure.
• Leaders must focus less on technology and more on setting clear strategic objectives and making tough decisions about value creation.

---

For business leaders and tech watchers, this episode is a grounded guide to understanding why so many AI pilots falter and what it will take for the AI revolution to move from “beanbags and slides” to mainstream business value.

Transcript

[A]

Foreign.

[B]

This episode of the Times Tech Podcast is sponsored by kpmg. Hello, I'm Katie Prescott and I cover technology for the Business Desk of the Times. And in this episode of the Times Tech Podcast, we are looking at how AI is really affecting companies on the front line. We know that businesses are investing heavily in the new technology, but many are finding that the hard part is turning early pilots and that investment into something that actually changes how the organization works, whether that's making everyone more efficient or seeing some kind of financial return, some kind of impact on their bottom line. So we're asking, is AI having an effect yet? Is it making businesses do things better and does it match up to the hype? Well, to discuss all that, I'm joined by Paul Henninger, who's UK Head of Technology and Data and and Global AI Leader at kpmg, and Professor Alan Brown from the University of Exeter Business School, who's the author of Making AI Work for Britain. Well, welcome to the podcast, both of you.

[C]

Thank you.

[A]

Great to join you.

[B]

Alan, can I start with you? So one of the big conundrums that I hear from organizations at the moment is they really want to implement this shiny new technology called AI and it's going to revolutionise the workplace. It's going to mean that none of us have to work, perhaps it's going to make everybody more efficient.

[A]

And.

[B]

And yet there's still this enormous gap between sometimes what they're spending or the dream of AI and actually what comes out the other side and the reality of it. And I just wonder, having written so extensively about this, what advice you give organizations when it comes to sort of trying to close that gap and be ready for the implementation of AI.

[C]

I think what we see often is that when people are looking to do something new, there has to be a compelling reason to act. For many folks, the pressures on them are just to get through the day. I want to do the things I do, I want to do them well, but I want to be able to get rid of the pressures that are on me to deliver the near term issues. The fact that we say to people, be more innovative, be more inventive, use new things, fail fast, fail early. They understand that pressure, but they have to come to that in light of the current day to day activities that they're involved in and the outcomes they're trying to achieve for themselves, for their organizations, for the people around them. And I think what we have to try to do is introduce some ways that they can do these new things and feed them into the objectives that they have on a day to day basis. And I think too often we separate the day to day business from changing the business. And we have this kind of idea that there's a small group of privileged people with the beanbags and the slides and the opportunity to do new things and there's everybody else keeping the lights on. And I think we have to start with the point of view of no, everybody is changing all at the same time, perhaps at different rates and perhaps with different risk profiles around the work that they do. But everybody is part of that change.

[B]

Paul, do you think that's the issue, I mean, from what you're seeing your clients at kpmg, that people are perhaps looking at this as a tech issue rather than more holistically as Alan's just described?

[A]

Well, I think they've started to look at it much more broadly for sure. And that's because it becomes pretty clearly apparent, whether you have a sort of theory about it or not, that there's a difference between technology change and business change on the one hand, so you can have, and we have seen AI doing things like automating 80% of financial planning or something like that. And that's an example of technology change. But if you don't adapt the business, the processes, the people, the teams, the governance around it, you haven't actually changed how the business functions. So you need to do that work. And then the interesting connection to value is that even if you change how the business functions, you haven't necessarily actually done something that's going to show up in the balance sheet. You have to change the inputs and the outputs. Either the cost has actually changed or the revenue's gone up or something like that. And just changing the way a business operates doesn't do all of the work to do that. And so it's definitely the case that even when AI works, the business doesn't natively change. You have to do some additional work and even past that in order to sort of capture the economic value, there's more work to do.

[B]

Again, what is the best example that you've seen? And you don't have to name a company if you can't. But what is the best example you've seen of that economic value being added or change being done successfully?

[A]

There's some amazing examples that we've seen of the change being done successfully. They happen to generally be in sort of parts of the business that are not stuff that I get excited to tell my mom about, you know, so in risk boring a little bit. But Yeah, I mean, in risk management, the introduction of AI has really fundamentally changed how that works over the last few years. Its ability to read a piece of legislation and analyze processes across an organization and identify the bits that a legislative change relates to and then kind of update a bunch of controls automatically, that's completely changed how people think about protecting themselves and implementing regulatory guidance. Now, that's, again, that's not going to get you on the COVID of a broadsheet or whatever, but it's really fundamentally shifted how business operates. Whether or not that translates to value is another interesting question. So that gets back to ultimately kind of a leadership question. So in the case that risk management all of a sudden is 50% more efficient, you've created capacity in the business. Risk management doesn't generate economic value by itself, except in a kind of protective way. So you have to decide, do we want to send more products out the door, which we can now do because we can manage the risk of those products. Do we want to reduce the costs of operating the business? And most companies, I think, relatively thoughtfully, are trying understand what is our objective. Are we trying to grow very quickly, are we trying to get into new markets? And they're starting to have the confidence that AI can deliver that capacity in order to make those decisions. But again, AI won't make those decisions for you. Ultimately, the leaders of organizations need to sit back and decide what is ultimately our business objective. And then we put the AI towards that end.

[B]

Essentially, you're saying if it's making one bit of the business more efficient, that might free up people to do more things or give you more money to invest in other areas. And so it's then how you direct that 100% correct.

[A]

And again, unfortunately, kind of boring example of risk management, it genuinely does free people up to do quite useful things. I mean, ultimately, if you're in a sort of financial crime department, you're better off focused on actual financial crime issues and not just wading through the detritus of anomalies and how money flows around, which there's lots of. So it has absolutely made, you know, risk organizations more effective. It's just in terms of how that shows up in gdp. There's a much more tenuous connection on some level.

[C]

Can I jump in? Because I think one of the big issues here is what I might call the value fallacy, which is, I think this idea of creating value and then realizing that value are two different things. And I think part of what we see in the figures right now is that people are seeing ways that they can apply AI and that may do some of the things that you just described, freeing up some time. But where does that get deployed and how does that get recognized is a different problem. So let me give you a very simple and perhaps a little trite example here, which is the government has said how much money they might save as a result of applying AI in the civil service. When you look at the figures, and I'm going to simplify just a little bit, the way they come up with the figure is they say roughly 25% of what a civil servant does as administration. That's amenable to AI. Let's work out the average cost of a civil servant, multiply by the number of civil servants. That's our saving. And you say, do you really think it's going to be that simple? How we free up time and energy from civil servants and that gets somehow banked as a recognizable value that we will see to the bottom line of the economy, or to create new jobs, or to hire new people, or to work with frontline services. And of course, it isn't that simple, but I think we still don't understand the way that that works, the economics of it. And I think we see that at the individual level. You know, I can use some AI to write memos and to analyze documents and to help me schedule better. And we can see it at the organizational level. I can do it to help me with risk management and in auditing and to help me organize my hr. But I'm not quite sure where that leads us as an organization to. To what that means for creating value, creating new economic opportunities and realizing that.

[A]

Yeah, I think that's true. I mean, there are other examples for sure. One of the early applications that we were involved in with large language models and a bunch of other technology was in a mortgage underwriting process. And it was a pretty aggressive re architecting of how mortgages were originated, basically. And the objective that the financial institution had was to get more data into that process so that they could grant more mortgages to more people or grant

[B]

the ones that they shouldn't be.

[A]

Yeah, for sure. But actually, interestingly, you could decide to do both things. You could make your portfolio less risky, or you could keep the level of risk that you have and have a bigger portfolio. And they chose the latter. And that is what happened. And it worked. But to your point, without setting that as the objective, if you just make the mortgage underwriting process 30% more efficient, you just have a faster process that's doing exactly the same Thing. So you have to ultimately decide, we want to deploy this to grow, which implies a certain kind of path that you take, or we want to deploy this to become more efficient. And the only way to bank the costs of, on paper efficiency is to reduce the number of individuals or materials involved in the process.

[C]

There's also a sort of lowest common denominator, weakest link kind of issue here because often these tasks that we can automate or improve or eliminate are linked to many other things. You know, the classic is the health system. You know, I went for a scan recently. Most extraordinary modern technologies around imaging, image analysis, the machinery we have, the way in which technology and software and analytics and information processing is brought together. And then it takes three weeks for somebody to look at it and they lost the paperwork and they send literally a physical letter to my healthcare system that eventually contacts me. All of these systems around us become part of the challenge when you begin to say, oh, look, we can do a 10x improvement in this one area, yes, but everything else doesn't have that ability to flex. And I think that's what we see in many areas, 100%.

[A]

And healthcare is really, and it has been over the last 10, 15 years, an interesting example of you have to be quite careful which problems you pick to solve. So one of the kind of coolest things people have been able to do with AI is to use AI to read MRI scans and other things and diagnose diseases. And it's incredible. It just so happens that that's probably not the most effective way to introduce capacity into the healthcare system. It is a way to do that. The most effective way that we've been able to introduce capacity into a healthcare system is simply by using AI to schedule the resources required to do the MRI scans to begin with and do things like check people out of the

[B]

hospital, make sure you get your letter on email.

[A]

Well, 100%. And the problem is when you get in a room and say, what are some interesting things we could do? AI, I could send out a bunch of letters, gets down to the bottom of the list, because that's not an exciting cool problem or whatever.

[B]

That's your risk management point.

[A]

Well, and also just the objective, if our objective is very clearly to introduce more capacity into the number of, it just may be the case that those kind of really fascinating, interesting mathematical problems end up lower down on the list, which is probably not a terrible thing.

[C]

And the big future path for this, which is where people are looking, is to say, let's take healthcare as the example we'd really like to say instead of using hospitals to fix sick people, let's prevent people getting ill using technologies like AI so that we can understand our current situation. We can self manage our health by looking at information from things on our wrist and what our dietary habits are and what our behavioral habits are and we can use those with personalized approaches so that we can understand how to keep you. Well, the problem we have is our systems aren't set up for that and shifting systems is much more difficult than shifting technology.

[B]

Well, perhaps that plays into my next question. I mean one of the big studies that's been done and repeated many, many times, right on this subject is the one from MIT which showed that 95% of AI pilots didn't work. When you look at pilots at KPMG and people investing cash in them, why don't they work? Is it the system's point? Is it the people point? What's going on?

[A]

The reason we are unable to or we need to do more work to get AI to scale is largely to do not with the technology it is to do with the data, but also with the people, the sort of financial processes and the governance. Who gets to decide? I mean I've seen pilots that don't get into production simply because it was unclear who could authorize that. So on some level, like I said, sometimes the problems are not that interesting. I have to say. I mean my reaction when read that MIT study was not like oh my God, 95% failure rate. It was like wow, how did they achieve a 5% success rate? Like most technology experimentation, a 5% success rate from. I have an interesting idea, I'm going to try something to I actually change something in real life that's like an incredibly high throughput on some levels. Obviously we shouldn't have things failing all the time. I like the sort of fail fast thing. I once said we should all get T shirts that said maybe you should succeed a little faster as well. However, if that 5% was something that could get people in and out of the hospital more quickly, then I don't really mind that the 95% of the ideas we had weren't the right ideas. The one interesting thing we've seen along these lines that is a little bit concerning I suppose is that the more successful the technology is, the harder it is to move it out of a pilot mode. Which kind of makes sense when you kind of look at the pilots. If you've automated something at a 10% rate, like I can summarize my emails for the day or whatever. It's not that hard to put that into your current process. I come in in the morning, I press the. But I read the summary of my emails, I go on with my day. If you do something like an 80% automation rate, you're talking about really, really fundamentally changing the course of hundreds of people's days and exactly how you kind of stitch all the rest of it back together is quite complicated. So ironically, the sort of more effective we've become at solving problems with AI, the more there is actual work and thinking to do for how do we adapt the process and put it back together sort of Humpty Dumpty style, such that it actually can be used in real.

[C]

I think one of the big areas we're learning a lot about where AI is being promoted as a major savior is in the software world creating new software.

[B]

Well, and some people are saying it could destroy the software sector, right?

[C]

Exactly. There's lots of application of AI in that space. But the data that's coming out is quite interesting from people like Ethan Mollick and others that are right at the forefront of this. We can generate a lot more code a lot more quickly using these technologies. It's not necessarily being deployed and used. And this brings us back to this, piloting the gap between being able to learn about technology, create new technologies and apply it in the enterprise scale world. And it's because organizations take time to consume these new ideas. You can't change things 10 times a day even if you can generate code 10 times a day.

[B]

Do you know what's quite joyful for me about this conversation is I spend a lot of time talking to tech bros and who talk about how AI is just going to completely change the world and it's going to eliminate all jobs. And actually hearing from both of you on the front line of implementing doesn't feel like that's the case at all. I mean, where do you stand on the hype around AI?

[A]

You know, I don't consider myself a tech bro, but maybe that's just a. I've never been a bro type, but I'm an extreme AI optimist. I mean we've been able to do absolutely incredible things with it, but optimist

[B]

in the sense that it' going to completely revolutionize society and mean people stop work.

[A]

That it has the potential to do that for sure. I mean the reality is that like that level of change is a lot of change. I mean there is a sense in which it definitely could be the next industrial revolution, but that significantly underestimates the amount of work that went in the first industrial revolution. I mean, people didn't just buy a bunch of steam engines and rock up to the place that people were knitting sweaters and say now where sweaters are different, like the whole idea, for example, a supply chain didn't exist. You used to have the sheep outside the place, you knitted sweaters and then you kind of shaved them and made a sweater or whatever. And then to make 100 times or 10,000 times more sweaters, you had to completely change how shipping worked and all sorts of other things.

[B]

Where do you think we are in the sheep shaving supply chain?

[A]

Well, we're doing things like buying steam engines and rocking up to a finance organization and saying, where do I plug it in? And the reality is there's a massive amount of work to do. So I think we are starting to see the sort of future happen in pockets, for sure. I mean, what we've seen happen in pockets in finance organizations and how products get designed and how code gets deployed is amazing. But for that to happen everywhere, all at once, such that that kind of sci fi future exists is going to require a lot of work and also require us to make some decisions about what we want that future to look like.

[B]

What do you think, Alan?

[C]

So I have some of that same optimism, but let me also reflect the other side. That sort of was what keeps me awake at night, as it were. I was part of the other revolution in technology in Silicon valley and elsewhere. 25, 30 years ago. There was significant impact of that change. Look at what happened with the Internet, cloud based technologies and so on. And there are two or three things at play right now that remind me a lot of those times. The first one is massive, what I will call overinvestment from technology providers. They've invested trillions, they have to see that returned. The second thing is that when people say to me, is AI a bubble? My answer is, of course it's a bubble, because we always overplay. But what's interesting is to ask yourself, what will happen when the bubble bursts? What will be left behind and what will that mean for us? And if we look at what happened 25 years ago, there was a massive bubble that happened at that time. What we got left with was incredible investment in technology infrastructure, particularly communications infrastructure and telecommunications. We got cloud technology. As a result of that, we got a massive new generation of people building software and systems in new ways. And we got a set of ways of thinking about the impact of the Internet and technologies on how we work. As individuals, as humans, how we collaborate, how we work online. And we had a cost base that because we over invested was very cheap. So we got a new generation of technologies, individuals, organizations, entrepreneurship around that technology which drove the next wave. I think we're going to see the same thing again. And unfortunately some people will get caught in the middle of that as they do in all revolutions of technology.

[B]

So having gone through that cycle, what is your and talked about the difficulties of the implementation of AI, what is your sense of when organizations might have it being used as standard?

[C]

We have to sort of get away from. You're not an AI company, you are an AI company. If you're using Microsoft 365 Office, you're an AI company that's built in. If you're using Salesforce, you're an AI company. Those things are becoming more and more built in. I think we're seeing our lives and our way of working shifting. And for some organizations it will shift fairly quickly because the environment around them will shift quickly. If I look in education, for example, in higher education, there's a lot of resistance to change because the environment around protects it. Regulatory. The way in which we create academic courses don't like change, academics don't like change, but the systems around them don't like change. In other areas that change has happened much more quickly. For example, in journalism, in media, in communications, some things have happened very quickly.

[A]

Yeah, I completely agree. I mean, I think it's important not to underestimate the extent to which the world is already really manifestly different in a way which has greatly exceeded the rate at which it changed because of the Internet, which I agree. I was also around for copywriting for example. Obviously changed within the first three months after ChatGPT was released. For better or worse. I've got some people that I work with that run a small property management company and they run their entire finance organization with agents basically. Now it's easier to do that than scale that to a 300,000 person retailer or whatever. So William Gibson, whether he said this or not, is credited with saying the future is already here. It's just not evenly distributed. And that's I think what we're seeing and I think the rate of change is unbelievably high.

[B]

I'd like to close by asking you more about that because you've done a report into five areas that organizations need to redesign for AI infrastructure, data people, financial measurement and governance. If you're a CEO, listening to this podcast now, or anyone frankly who's involved in implementing AI in their organization. Where do you start and what do you advise people?

[A]

So I think all five of those things are important. The way this technology works, even just mathematically, is by setting an objective. You define an objective in a lot of detail. You put some algorithms against that objective and some data and it tries to figure out if it can solve it. And sometimes it can tell you the members of the Kardashian family or whether a credit card transaction is fraud. But it works a lot of the time, but it doesn't work without an objective. Ultimately, the pressure and the opportunity is on leaders to decide how do I want to change my business? And until you decide how you want your business to change, the AI, just again, even at the level of mass, doesn't know what to do. It can suggest some options and can be used as part of the research. But the defining an objective down to a level of detail, which is actually probably pretty uncomfortable because it will involve trade offs of profit versus growth and people versus technology and all kinds of different things, is absolutely the most important thing to be spending time on.

[B]

Thanks so much, both of you. That was really fascinating. Well, my thanks to our guests Alan Hellinger and Professor Alan Brown. This was a sponsored episode of the Times Tech Podcast brought to you by kpmg.