#85, KPMG’s UK Head of Connected Technology and Global Head of Digital Lighthouse, Paul Henninger: Is AI making or breaking the jobs market?

0:00One of the sort of frustrations that people have had with digital technology is that although it's resulted in productivity improvements, we still are working as many, if not more hours as we used to, and it hasn't freed us from hard work on some level.

0:19He's in his second stint at KPMG and very much at the forefront of the AI revolution. We attempt to unravel the layers and explore how generative AI is evolving and fostering a seamless human-machine collaboration. We delve into the ethical considerations, macro trends in AI that Paul is seeing, global adoption, and the gold rush for AI productization. And we also put a spotlight on the societal impact and possible job displacement, amongst other things. You really do want to miss this one. It's a great episode. It's Paul Henninger.

1:02Paul, thank you so much for coming on the Tech Leaders Podcast. I'm really, really excited. Do you know what? This is the first time we've had a second guest from the same organization onto the show. So congratulations. You're a part of the Tech Leaders Podcast first. We had Lisa Hennigan on probably about a year ago now. And this is the second person from the almighty KPMG, one of the most well-known global consultancy firms in the world. There's so much to talk about at the moment, Paul, with this explosion of artificial intelligence. But let's start with the immortal first question of the Tech Leaders Podcast.

1:35What does good leadership mean to you, Paul? It's a good question. I think what I've come to believe is that good leadership requires, as many things do, a strong and sometimes embarrassing sense of curiosity. I find it impossible to understand all of the details of the technology teams that I lead today. But I also find that I'm a much more effective leader when I'm willing to ask them a bunch of stupid questions and then progressively slightly less or more interesting questions to understand enough about the problems that they're solving, the way the technologies work or don't work, such that I understand the context that they're operating in.

2:15Once I do that, then it's some of the usual things. I never ask anyone to do something I wouldn't be willing to do if I had time. And I think once I understand enough about what they're doing, it's a lot easier for me to give them the freedom and support and challenge that they need to do it really, really well. But the starting point is that curiosity about exactly what the heck it is that they're going through, struggling with, researching what's next for them. And once I understand that a little bit better, albeit not as well as they do, I find it, you know, decent leadership to be within grasp.

2:47Yeah, for sure. I think retaining childlike curiosity as you get older and more experienced, it's quite a challenge that is, isn't it? Because you kind of develop a bit of a, not an ego, but you feel like you shouldn't be curious anymore almost as you get older. But you definitely need to be, because if you're not and you become too proud to ask the stupid questions like you refer to there, then it probably has a detrimental effect on your own development and your own leadership skills, doesn't it? So tell us a bit about yourself, Paul, from the point of graduating from Princeton all them years ago through to where you are today.

3:28I got my first computer, or I got access to my first computer when I was five years old. So I don't think anybody in my generation is considered a digital native in any technical sense, but I've been trying to learn how to code. What computer was it? It was a TRS-80, which was one of the first laptop-looking laptops, and my dad brought it home. And I don't actually remember where I got a book on basic programming, i.e. beginners, all-purpose symbolic instructional code. But I got one and started to write, you know, line 10 print, Paul, line 20, go to 10, and then various other things from there.

4:05But I've been working with trying to make computers do something fun, I suppose, originally, and then interesting after that, and then useful after that for a long time. Ironically, though, I studied architecture, and sometimes I just leave that as a word, and people think it's network architecture or something like that. But it's the, you know, the buildings. The old-fashioned kind of architecture. Yeah, exactly. And there was a project that I did, which is how I kind of got firmly ensconced in the world of data and analytics. And I suppose ultimately AI, where I and a couple friends took some video from closed circuit televisions around the world, borrowed some software from a physics lab, converted the video into these particle flow models and looked at the interactions between pedestrians and cars and other objects.

4:49and at the end kind of printed out the design for a funny-shaped park bench. And I remember there was a specific moment during that effort where I thought that the technologies that underlie the design for our funny-shaped park bench might have something more useful for the world to offer than park benches. Not to say I don't enjoy a good park bench. And so I sort of set off to try to find where I could learn more about that stuff and understand how it was going to impact the world and ended up through a succession of software companies, including Fair Isaac, and have been working ever since then to try to learn more about what I think of as data-centric technologies, although most technologies these days are data-centric, and try to figure out what problems they might usefully be pointed at, try to assemble groups of people, usually with relatively diverse backgrounds in terms of how they think about things around those problems, and see if we can identify enough value that we can keep going with it.

5:42And some of that's gone really well, and some of it's, as you might imagine, been a few dead ends. And then 25 years of that, basically, more or less. Oh, that's fantastic. So you were sort of playing around with things that would lead to AI, essentially. I don't know what you would call it in the early days. You were in there really, really early, I suppose, then, yeah? One of the things that I ended up working on, this was about 14 or 15 years ago now, was using machine learning to do something which everyone's quite excited about now, 15 years later, which is to detect cancer or diagnose cancer from MRI images.

6:18And the interesting thing was we, from a technical point of view, I and the group of data scientists and engineers had worked on that at that time, succeeded in producing a model that could identify a certain type of cancer as well, or I suppose more accurately better than the panel of doctors that we were using to compare the results to. And it was an interesting failure, mainly because technically we succeeded. We got something that was right more often than people, but we hadn't thought a lot about the very human process that surrounded a cancer diagnosis and kind of more to the point, the even more human process that surrounded not just the diagnosis, but the sort of entire experience that someone has from something not feeling right to go into a medical professional and have a set of meetings or discussions and then getting a diagnosis and then in the sort of worst case going through a treatment and recovery and post-treatment and all sorts of other things.

7:12And there were two other projects that I'd been involved in in the intervening years, which got better at understanding those very human problems and starting to think about how artificial intelligence got built into those human processes. One of the main lessons that I have from pretty much every advanced analytic project that I work from is that although it's hard, the easy part is getting the analytics to work. And the hard part is thinking enough about and in an innovative way, enough about all of the sort of, you know, human decisions, human cycles, the speed at which humans interact, think, do their work, to figure out how the incredible speed with which a machine's capable of analysis can be synced up with a human process.

7:56Yeah, fantastic. This episode is brought to you by Be Digital UK. Be Digital are a trusted partner to leadership teams of technology-driven organizations delivering ROI-led solutions across cost optimization and also data and AI adoption. Looking back now over the course of your career, what were the key milestones, Paul, that you think set you on the path that you ended up pursuing? What kind of things stand out? One of the key moments was literally physically moving, you know, which was not a particularly digital experience that involved lots of boxes and airplanes and things like that. I moved from New York City to London about 10 years ago.

8:39And it required that I not relearn things. I knew how to make a sandwich. I knew how to take the tube. I knew how to turn on the television and all sorts of other things. But it required that I re-sync with a kind of culture and just a set of very subtle differences. And that was an opportunity for me to a little bit re-evaluate some of the assumptions that I had that I didn't even realize that I had, I suppose, at the time. I had had the observation at the time that the set of technologies that I worked with were advancing at a rate such that the problems that my customers or clients had and were bringing to me exceeded the rate at which I could respond to them, you know, in any version of any software company, whether it was literally my own startup or very, very large companies I worked for.

9:27And so I partly, after that move to London, ended up working back in the consulting and or advisory industry, which, you know, to my assessment was, you know, the most logical place to lean into the proliferation of problems that we could potentially solve with, you know, again, data centric or other technologies. And so that was a critical moment. I did my own startup, which is to say it was, you know, I was tied for employee number one with a friend of mine. That was an incredible experience. I mean, it was ultimately kind of successful in the end, you know, in terms of how those things are measured.

10:05I thought I was willing to kind of do anything that needed to be done. I'm always one of those sort of, to the extent that I'm a leader, I'm definitely a servant leader. There isn't anything that's below me, even though I do know that there's better uses of my time and worse uses of my time. But when you literally are starting from an actual blank sheet of paper in terms of what problem you're going to solve, how you're going to approach solving it, how do you pay employees, where do you have an office? It was an incredibly formative experience, as well as the experience of not having any brand or brand permission at all.

10:37You know, working as I do now at KPMG, you can, you know, people know who KPMG is. They don't wonder why we're there to talk to them. When I was at my startup, people knew who I was and some of them were willing to talk to me, but they'd never heard of what we were doing or why we were doing it. It was relatively new and novel. And so it was a fascinating experience to have to think through how do you connect with people with problems in a different way and think about how you can help them and get them to take a little bit of a bet on your ability to execute in a way that was different than anything that I've experienced before.

11:07Yeah, I think the selling part of it takes on a whole new challenge, doesn't it? Because nobody's ever heard of you. How did that startup situation, because you worked for a relatively stable company, you had a really good role. Was it Actimize? Yeah. You were the VP of products there. And then all of a sudden you start your own firm, I think, around the time you were leaving. And then go it alone for the best part of three years. How did that come about? And what were the key lessons, apart from what you've mentioned? What were the key things you learned from that experience which stood you in better stead, obviously going into the role with BAE and beyond?

11:41Like a lot of things in my life or in my work life, it did start from a kind of technical observation which stuck in my head, basically. Actimize was a company that developed analytical solutions to detect fraud, to combat money laundering, and to help firms with compliance problems in capital markets. And so we had some advanced analytics and some rule sets and all sorts of other things. And at the time, everyone was interested in the sort of problem of explainability as they are today with AI. And the sort of perception was that sort of advanced analytics or machine learning was not explainable, whereas rules were.

12:16And so I had an argument with people to explain that rules were very explainable, but other things were equally as explainable. But my observation was that people were really, really into rules and had an idea about using complex analytics to break rule sets apart, optimize them, and then reassemble them as even better rule sets, basically. And so eventually did work with some folks to come up with a modified genetic algorithm based approach to doing that. And so having identified a kind of literal solution in search of a problem and having been convinced that the solution was interesting enough and applicable enough, you know, also just ran into a kind of, you know, a normal, you know, cycle of personal achievement or whatever.

12:58We'd grown Actimize from a kind of 30-person company with a couple million dollars in revenue to a few hundred-person company with a couple hundred million dollars in revenue. And so it was just an interesting inflection point and took the opportunity to take this technical idea we had and go in search of a couple problems to solve, you know, set up the company. And actually, interestingly, initially identified people analytics as a place to solve it based on the observation that that was the one area of a modern enterprise that in effect used almost no data or analytics to run things. But again, starting point, as it often is with me, even if I'm no longer the one writing any code, I get quite interested in the potential of technical ideas, whether they're big ideas or small ideas or simple ideas or complex ones or whatever.

13:48And my passion has always been, as it was at the beginning of my career, to see what can we find to do with this. And that was interesting enough to build a company around in essence. Yeah, sure. And did you sell the company, I'm assuming, then? We did, yeah. We actually sold it to a consulting firm. And actually, the story of my life is consulting firms trying to become software companies and software companies trying to become consulting companies. There's always, for whatever reason, it must be something about me. I've always managed to get involved in the tension between how do you take intellectual assets and form them into software of one type or another, or how do you take intellectual assets or technical assets and form them into the services that people need to make technology useful.

14:34Obviously, you would expel a BAE, KPMG, you went to another consultant firm and then back to KPMG. So what made you go back and tell us about your current role as head of Digital Lighthouse and what exactly that means? KPMG, I mean, it's a bit of a cliche, but I came back to KPMG for the people and the culture. The thing that had attracted to me the first time around had been a group of people that I knew who started working there. And the thing I loved about it was the way in which collaboration wasn't just a kind of hard to argue with way to get things done, but was part of the literal DNA of the organization.

15:13That was why I came back. And it's one of the things that makes me excited to come into the office or to press a button in my home office and become part of work every morning as we do these days. In my role leading Digital Lighthouse globally, on some level, Digital Lighthouse is something very simple, which is it's the global AI analytics and emerging technology center of excellence. And what we do in that regard is connect, set the strategy for and help build capability around the world for KPMG in those areas. AI and analytics being relatively straightforward, technical domains, albeit ones that everybody's interested in these days in a relatively intense way, and emerging tech being a relatively constantly shifting things, you know, stuff we're focused on these days are things like quantum and even still metaverse despite the slight drift in people's interest in that.

16:04But what we found over the years is that the capabilities that you need to make those things useful are more than just AI and analytics. And so we started delivery networks around cloud and testing and automation and all the technologies that need to be assembled to deploy analytics into the real world, into a business, into a government agency or department or ministry to solve a real problem. And ultimately, what we noticed on some level that we've gotten good at was building digital things, whether they're products or solutions or assets that consultants use to do the work that they do. And so in some respects, Global Lighthouse are the sort of digital builders of KPMG, who partner with all sorts of parts of the firm, definitely in consulting and advisory, but equally in tax and audit to figure out which of the technologies with which alliance partners and what configurations can be brought to bear on how we solve problems for our clients across domains from financial crime to internal audit to global business services and finance to slightly more arcane model risk problems and capital markets and all kinds of things.

17:17It ends up being the thing that I've looked for my whole career, which is the opportunity to assemble an incredible expert team, in this case distributed around the world, on a set of really interesting technologies, but do so only in the service of working closely with people with domain expertise in some specific part of a business problem or operational problem or process problem or whatever, and figuring out together how we might build something that can solve that better, faster, cheaper, or in a new way. Fantastic. Okay. So, well, yeah, well, thanks for clearing that up. I think it was quite an interesting job title, so I wasn't sure what was meant by that, but that makes total sense.

17:59So let's talk about AI, Paul. I think we were all quite blown away with ChatGBT around about this time last year. And since then, it seems to be growing and going from strength to strength in terms of these large language models. And it felt like AI capability had finally come to the masses, didn't it, with ChatGBT. But there's use cases popping up everywhere. There's tools popping up everywhere. There are concerns and maybe justified concerns about AI displacing jobs. In your view, does AI primarily lead to job displacement, or do you see it as a tool that can also be used to create jobs? I see AI as fundamentally a transformation technology or transformative technology, because I do think, well, mainly because I've seen the impact it can have when it's configured and deployed in the right way.

18:48It can fundamentally change how we do something, whether it's prepare an argument in the context of a litigation, whether it's conduct a sanctions monitorship, whether it's price an ad block or something like that. And I do think that there are decisions. One of the things that AI introduces is significant productivity, although it's not the only thing. And when you're able to generate dramatic productivity improvements, a three or 5 % productivity improvement is dramatic enough. If you look at the sort of estimates that we've done and that others have done over the impact that AI could have from a productivity point of view over the next five to 10 years, you're well into the sort of 15, 20 % impact.

19:30And when that happens, we have a decision to make, as you do with any kind of productivity improvement. And that is whether we do more with the sort of labor that we've got, or whether we decide to sort of pack it in and kind of keep the same amount of output and problem solving, you know, with fewer people, which would mean there would be displacement of the actual people doing work. And I think what's more likely is that AI will stimulate the type of growth that we've seen in places like the Internet. I mean, one of the sort of frustrations that people have had with digital technology is that although it's resulted in productivity improvements, we still are working as many, if not more hours as we used to.

20:10and it hasn't freed us from hard work on some level. And despite being at the sort of tail end of, let's say, 15, 20 years of the last round of digital transformation, we still have job shortages in all sorts of places, albeit some interesting economic impact of that stuff. So I think we're likely to see something very similar. I do think you'll see a continued change in what types of job exist. People used to write things down and erase things from and sum things up in ledgers. And then someone invented the spreadsheet and those jobs completely went away over the course of about a decade. And then in their place, you know, the analyst role popped up and the sort of line and the growth of that job literally mirrored the decline in the number of bookkeepers.

20:55And I think you'll start to see something very similar, which is to say that there will be jobs where AI automates a very specific set of tasks that we have to do. And one thing we will have to pay a lot of attention to is, how do we help manage people through that? If you've done the same thing every day for 15 years, it's not just as simple as clicking on a learning module and finding something new to do. And so a big part of how the transformation will occur that AI offers is as much the people processes and support for skill-based careers and the evolution of careers with regard to skills that kind of get clustered into roles, that get clustered into jobs.

21:37That started to change in a big way over the last decade. And I think we'll see an acceleration of that. So I don't expect that AI will result in, it'll result in a major, major change in labor and an acceleration of some of the things that we've seen. But I think it's much more likely that we see more things for people to do rather than less over the next five to 10 years. But again, I think my experience of how that stuff has played out is it's not like, you know, we go home on a Friday and then on Monday, we don't need tellers anymore. You know, it's even the rapid versions of those processes take months, if not years.

22:13And so, and part of the planning that needs to go into all of that occurring is, you know, what is the impact on the people that we have doing that stuff and what else do we need them to do? Because I do think that the sort of cliche around once people are freed up from having to do very rote, very mechanical, very repetitive skills, there's lots of other things that we need to get onto that require the capabilities that humans have that machines don't yet have and probably really won't even if the machines are getting more clever and more intelligent. Sales type roles, commercial roles, negotiation, those type of things that surely they're impossible to automate, aren't they?

22:51But if you're doing some sort of process disorientated role then i think you're probably at risk aren't you because uh the team of 10 can probably now be replaced with a team of three if that makes sense you know which is quite scary but i mean again i think yeah as always history shows that new new jobs new new ways to contribute to the to the collective will emerge i'm sure so uh are you seeing any any macro changes at kpmg at the moment. Obviously, you've got a large client base. Are you seeing any shifts already? Because this is still really early days, isn't it? We're not really seeing AI being adopted by organizations and employers too much just yet.

23:33But I think that's going to change in 2024, and even more so in 2025, I'm sure. But are you already seeing macro trends emerge here, Paul? Yeah, absolutely. I think we're seeing relatively fast adoption of this current generation of AI technologies, but particularly in areas that have a couple characteristics. One is that require assiduousness, which is a very sort of 10-pound word, which fundamentally just kind of means quality repetition, and that are relatively insulated from the litigation that's occurring around the intellectual property disputes, which are still very much in progress around the creation of the large language models and their siblings in generative AI.

24:24And, for example, we've seen really strong and really rapid demand for the use of generative AI and the set of other AI technologies in areas of risk. You know, not all risk problems are exactly the same, but fundamentally, you know, in certain parts of it, you get external risk, which is usually sort of described, you know, officially in some type of law, which then a, you know, corporate entity has to describe as a policy. It sets up some controls, needs to test things to see if they meet the controls, and then usually have some sort of quality assurance process. And that kind of six-step process turns out to lend itself incredibly well to enhancement and automation via the use of generative AI.

25:04And partly because we've been working on that for close to a decade, we've seen the sort of acceleration that occurs as a result of generative AI relatively intense. because I think that is an area where there's lots of mechanical processes which are done by people. And there's lots of really complex things to think about with regard to what's our actual risk position. You know, what are the external changes in markets and regulation and the world, the climate, etc., where people would be much better off if they were redeployed to think about those fundamental risks if they didn't have to just process the day-to-day work of risks.

25:40That's one area that we've seen quite a lot of. Okay, that's interesting. The other thing we've seen significant progress and interest around is the use of AI to support people processes, i.e. the HR function. And again, it's a simple set of use cases, but it's just more about taking advantage of the conversational, the very human kind of appearance that generative AI has to help translate between the sort of real life questions that we have when we have a question for HR and the sort of kind of complex and technical. environment that actually is the day-to-day existence of HR professionals. And so again, you have a set of policies and procedures and rules for parental leave or holiday or being able to buy or trade holiday days for compensation or all this other complex stuff, which you need very detailed processes for.

26:31But when I run into a situation where I actually want to access that, it's not because I'm thinking about it in policy terms. It's because my partner just said to me, What if we took three weeks off next August instead of two weeks or something like that? And I need to sort of figure out if that's even possible or what I would need to do. Is there a form to fill out? So that translation capability between a complex corpus of information and real life is something that it turns out there's a lot of demand for and it's easy to start to deploy. There's a number of areas which have already kind of taken flight.

27:03And then a whole other set of others, which are actually the use of AI in implementing technology, you know, the sort of turning the technology back on itself, you know, so AI can be incredibly useful in configuring an ERP system or in testing something. And that's an area where I think there's a lot more work to do to make sure it works because you can't just, you know, kind of ask the AI to implement your SAP platform and assume that it must have done everything right because it kind of looks like it works. You have to be very careful about that. But that's an area where I think very quickly we'll start to see some major change in the work that we do.