Why am I not seeing value in AI?

Jenna: Welcome to the Insight in 15. My name is Jenna Glass and I'm joined today with Paul Henninger, KPMG's Head of Technology and Data in the UK. Welcome to the podcast.

Paul: Good to be here. 

Jenna: AI continues to be the talk of the town.

Started maybe in the early 2010s, where we started to observe the benefits of basic machine learning and basic task automation with the use of chat bots. Then we entered our generative AI era where we are creating content and using gen AI for marketing and, you know, writing happy birthday texts to our friends and family - maybe that's just me. But now we're looking to operationalise it and scale it. 

But I think companies are slow to see the value. Can you talk a little bit about why that might be?

Paul: Yeah, I think what we're seeing is a good example of the phrase, the future is already here, it's just not evenly distributed. As a sort of rule of thumb, the best case scenario we've seen is that companies are able to take about 25% of the automation that they're able to create and turn that into actual balance sheet value. 

So for better or worse that means if you can automate an end-to-end process about 40%, that only means only 10% of that value is releasable. So there's pockets of value. In a sort of local sense, it's pretty amazing. But we are definitely seeing that companies are still just about to work through a lot of things, a lot of change that needs to occur in order to close the gap between that 10% and 40%.

Jenna: So when you say value, maybe we should take it a step back. What do we mean by value in this case? 

Paul: In practice we're seeing some amazing things. I mean in the use of AI tools to do engineering in our own work to build AI.

Ironically, we're seeing a 10x increase in the rate at which we can build things, compared to what we could before, and what we could do before was we thought pretty good. In other places, like in an FP&A in a finance reporting process, we've seen 50, 60% automation, lots of parallelisation going on such that we can do lots of things at once that we used to have to do in order. 

In really technical tasks that we've done, like with financial institutions where we have to assess their credit models once a year, we've seen work that used to take three months and very, very technical kind of, you know, regulatory, very sensitive work, we've been able to automate that at a 30% rate. So in very meaningful ways, we've seen real changes to the speed and quality with which we can operate all sorts of different parts of our organisation.

Jenna: What are the barriers that organisations are facing when pursuing, you know, the real value of this technology? 

Paul: We're pretty consistently seeing the same thing as people start to introduce AI, and particularly as they start to scale it. And the three things they encounter are fear, focus and friction. In essence, what we found is that AI as a technology has matured incredibly quickly.

I mean literally every week that goes by we have some new capability, we have to redo the architecture. It's amazing. And it's incredible what we're able to achieve now that we couldn't even just six months ago. But as the organisation starts to absorb this technology what we found is that it has to change in all, any number of ways. 

And that makes sense. You know, if this is meant to be the sort of next industrial revolution, we can look at the last one and the amount of change that occurred over a 25-year period as people went to, you know, process based manufacturing, as they invented the even general concept of a supply chain. I mean, the world became unrecognisable.

The same thing is happening now. The sort of people processes we have, the governance, the decision making, what our jobs are, what it means to manage an organisation, all of that's changing on the fly. And the fear factor, you know, the focus that's required to do that, and the frictions that start to come out when you change everything all at once, are the things that are the fundamental barriers. 

And what it points to is that this is as we thought it could be, a real transformation problem, not just a technology problem.

Jenna: I love when we were talking earlier just that phrase of, you know, this is not your typical change management problem. This is the next industrial revolution. So really approaching it with that mindset and how you tackle it is going to be unlike, probably, anything we've done in modern, recent times. 

Paul: Yeah, absolutely. And you can imagine, I mean, most of us, for better or worse, can't just press the pause button on the organisation we're trying to help run. So in addition to the fact that we're sort of creating a new way to manufacture the things that we manufacture, whether that's a loan or a sandwich or, you know, a birthday invitation, like you said, we're doing that while having to run the existing factory.

And that's one of the real challenges, you know, the sort of focus problem we have is that every sort of move towards a new operating model that's built around AI is a bet in certain respects against business as usual. It's not to say that, you know, that they're in opposition to each other, but you're running two things in parallel, and that's one of the most complicated forms of change in, you know, that there is - having to sort of repair a process, stitch it back together. 

It's hard enough as it is, but we're really embarking on something that's intended to be a really significant challenge. And then, surprisingly, almost every organisation that we're working with, if not every single one, is finding it challenging.

Jenna: So fear, focus, friction. Some beautiful alliteration there. Let's start with the fear factor. What can you tell us about that? 

Paul: I mean, there's initial fear anytime something new is introduced, you know, what does this mean for me?

How are things going to change? What we've found, though, and what I found actually, is that as you start to get into it, you know, AI can really, genuinely help you do what you do better, differently, faster, in ways that you didn't imagine. You know, I'm, you know, vibe coding applications while I'm sat, you know, watching my four year old daughter do swimming and stuff like that. I'm paying attention to her, but, you know, doing a little prompt and then going back to watching. What I realised at a certain point is that there's a real fundamental sort of fear that comes from the fact that this technology can do something that's really core to my identity. You know, it can, it's amazing in helping me think through strategies, how I'm going to approach, you know, operational decisions and stuff like that. 

And you start to wonder, what's my role on all this? This is literally what I've spent 27 years of my career learning to do. And all of a sudden this thing can really do it, you know, in a way that's completely different. And I think that's fundamentally the fear that we're all encountering. It's, it's almost a psychological safety question in a way.

If this thing is, you know, fundamentally changing everything that I used to attach value to, what does that mean for me? That can be a very productive question and experience, and pushing through it and finding ways to sort of talk about it, work on it, and turn it into a programme of change, is immensely productive. But it's genuinely a kind of, frightening experience at a certain point. 

Jenna: Yeah, I'm confident we'll, like, figure it out at some point. I don't know if we know what that looks like yet, but for myself, that's what I'm thinking.

Paul: Absolutely. We'll get there. 

Jenna: So moving on to focus. How can you do that and effectively manage the change brought on by AI while continuing to run your business?

Paul: What we have believed from the very beginning is that putting your best AI engineers, your best business operations people on the change is the way to get the best outcomes most quickly. 

In practice, that can be a hard thing to sustain. It's fine to say we're going to do an interesting three-month project and we'll see what happens. But to say for two years, I'm going to sort of take my very, very best people and put them on something that, that is not what they do every day for a living can be very, very difficult.

But what we found in practice is that that is in fact the way the leading organisations are creating the most value. If you want to change the way customer onboarding works or the finance process, or you want to create a new way to interact with customers, having the top folks work on that while the rest of the organisation operates the business in the way that it has to operate, can be achieved. 

You can achieve those two parallel runs. It's very difficult, but ultimately it's about making sure people are focused on the right things.

Jenna: So obviously we need controls to help manage this, and that's where the friction comes in. So how are organisations controlling, effectively controlling the roll out in the use of AI without slowing business?  

Paul: I mean that's interesting - I tend to avoid like deep technical, you know, answers. But in this particular case, there's a lot of progress being made and still some work to do on how do you instrument, you know, the sort of observation of all these little intelligent robots scuttling all around our organisation doing things like filing financial reports?

The sort of, what we tend to think of it as explainability or observability, and people are genuinely starting to make progress on how do you keep track of these things and identify not just when they hallucinate and do something that's demonstrably terrible, but how can you tell when they're starting to behave in a way that's not as you intend? 

And then you go back and, you know, and re-instruct them or something like that. So people are actually making progress. Unfortunately, there's no kind of click here to fix it solution yet. But generally the frictions that this is creating is not just a technical friction. As we start to introduce that level of automation into a finance process, you can imagine, I mean, finance processes are supposed to work the same way every time.

Ideally, they're, you know, designed to deal with anomalies and things like that. But mostly you don't want, you know, people kind of freelancing or improvising your end-of-quarter close process. And if you introduce a 40% automation rate, it's the equivalent of kind of, freelancing it in a new way. And so there's all kinds of reasonable processes we have and people and ideas and experience that pops out of the woodwork to keep that change contained. 

And that is a meaningful barrier to progress. It's probably a good barrier to progress. And the only way we've found to sort of problem solve it is to create the change, find out how the, sort of, immune system responds, and adapt that immune system in order to do the change in a controlled way. There's good technical solutions emerging, but ultimately it's a question of process, people, of risk tolerance, and all the things that we've always had to think about as we navigated, you know, turbulent waters in any number of respects.

Jenna: Yeah. So the human aspect is going to continue to be, you know, a key part of our AI journey. And we talk about human in the loop. So as companies scale, as they, you know, grow their AI capability enterprise wide instead of just, you know, one person using it to proofread, you know, work product or whatever it may be. 

How are you seeing organisations effectively use the human in the loop throughout all of this?

Paul: I think the key where it's working is kind of changing mindset from 'human in the loop' to 'human in the lead'. Human in the loop, it's a good phrase, and you think of a loop of a process and where do the humans fit in, and it makes sense. But equally, if I told you, I'll keep you in the loop, it's not exactly a kind of vote in confidence that I need you for what I'm doing. When we think about human in the lead, what it means is the AI doesn't know what to do by itself.  

There's no agent model, you know, that can kind of sit there and anticipate what we want, what our objectives are. We need to give it an objective. Mathematically, that's how the technology works. And so the human in the lead is all about defining the objective and identifying the places in a process where humans are critical to achieving that objective. There are definitely things that humans are much, much better at. The judgement, the discernment, making ethical decisions, you know, choosing between two or three paths or something like that.

Humans are excellent at that stuff. And designing an AI system or an agentic system is not just about arranging a few different robots in the right order, or with the right orchestrator, it is about defining the role that humans can play in order to re-articulate and to make selections and discernment in order to achieve the objective. So that human in the lead model we're seeing makes a massive difference in terms of how far you can go. If you just sort of check what the AI does at the end, whether it's developing an application or filing a financial report, you can make a little bit of progress. 

But if you run things that way as opposed to with humans and AI working in tandem in a really intentional way, you just aren't able to achieve as much. And so, you know, the right model we're seeing isn't just safe, but it's how people are really achieving effective change and starting to release the value that the technology promised from the beginning.

Jenna: So, we're almost out of time, and one question we like to ask every guest is, what is one key takeaway that you want to impress upon our audience in terms of seeing the value, effectively scaling AI in their organisations? 

Paul: I've said it before, and I think the key is we need to genuinely approach AI value as a transformation problem.

That means there's people issues, there's finance issues, there's governance, and genuinely a set of technology and data issues that need to be solved. The pilots, just like AI doesn't know what to do if we don't give it an objective, an AI pilot doesn't know that it needs to become a scale programme. And so, so the key is to define the objective, you know, understand this is going to be a lot of change and that change has different dimensions, and start to set up a programme. 

That's the way that we see top organisations that we work with are just that we talk to really, genuinely starting to make change that's excellent for everybody involved, that produces good financial results, but also genuinely creates new opportunity to create value in new ways: new products, new markets, new customers. And that gives us the real sort of, not just the financial value, but the really fascinating opportunity to, you know, see what's possible with a really powerful technology.

Jenna: Well, that's our 15 minutes. Thanks again Paul for joining us. You can read more of Paul's insights in a report published later this week called Make AI Scale. And you can catch all of our episodes on Apple Podcasts, Spotify and YouTube.  

Thanks again for joining and see you next time in The Insight in 15.