This transcript was generated using AI and may contain errors. Please check against the audio before quoting.
Abbie Day 0:04
Hello and welcome to the In the Loop podcast. My name is Abbie Day, one of the reporters at Insurance Insider, and I cover all things cyber, AI and insurtech. As part of my job, I have really fascinating conversations with market leaders about AI and its role in the insurance industry. And now we're bringing those conversations to our readers and listeners. So listen in to new episodes fortnightly and join me as I dive deep into the fast-moving world of AI, talking to tech leaders and senior industry professionals about how this technology is changing the London market and how you can use it to your best advantage.
In this week's episode, we set the scene with Rachel Turk, Chief of Performance and Strategy at Lloyd's, to get a feel for where the London market is positioned in the AI race.
So we are kicking off our podcast with a very exciting first guest. We've got Rachel Turk, Chief of Performance and Strategy at Lloyd's, joining us. Hello, Rachel, and welcome.
Rachel Turk
Hello, thank you for having me.
Abbie Day
Thank you for coming. I thought, given it's our first podcast, we could take a deep dive into all things AI and tech in the insurance market, particularly in London. So where is the P&C industry, particularly the London market, in the AI race? Has it had a good start? Are we already behind?
Rachel Turk 1:20
I think there's a lot of positive action in the AI space in the market. I would say if you put us against personal lines, probably behind. But is the London market behind other P&C? Absolutely not. I think it's just easier to see how AI can get deployed in personal lines. There's a lot more efficiency savings, it's much more standardised products, it's easier to deploy AI.
We are seeing a lot of innovation in the market now. We're seeing a lot of trials, a lot of tests, a lot of pilots. You look at Lloyd's overall, it's been around for over 330 years. That would suggest a strong history of innovation and constantly reinventing the business model. So I have a lot of confidence that the P&C market will reinvent as AI deployment comes through. But I guess broad scale deployment, like deployment at scale, I haven't seen so much.
Abbie Day 2:10
So how would you score its efforts? How's the market getting on?
Rachel Turk 2:13
So I think effort-wise, really high. There isn't anybody I speak to that is not thinking about how to deploy AI through their business. In terms of achievement, it's probably a little lower. I think every pilot that I'm seeing, it's just a bit harder than people imagine to make it truly revolutionise in the way that they would like.
AI can then allow you to put the analytics on top of it. So I think you're allowing some companies that are slightly further behind in their transformation journey to now miss out that step. I think that's great because it means the industry will now be able to move forward more quickly.
Abbie Day 3:12
Okay, and what's making it slightly more difficult than people are expecting?
Rachel Turk 3:15
Well, one is there's no lack of ideas. So lots of things that people want to do. One of the challenges is actually prioritising down that funnel of going, okay, we've got a finite amount of investment dollars and also, because actually the tokens are quite expensive, what are we actually going to focus our time and attention on?
And then if you think about how AI can be deployed in the insurance market, the initial easy use cases are efficiency gains, especially on the underwriting side: the underwriting process, the claims process et cetera. Standardising everything that we have and standardising the product so that AI can run over the top is actually harder than you would imagine, because the nature of the Lloyd's market and the London market is that it is bespoke, client-led solutions.
We always find a way to say yes, as opposed to something slightly more in the box, maybe take the US admitted market, which is going to go, here's a rating scale, this is what it looks like, and if you fall outside of it, the answer is no. Whereas when we try to find a way to say yes, it's actually quite hard to take that human wealth of experience and train an AI agent to be able to think in that way. And that, I think, is what's making it slightly harder to deploy. So you're seeing it deployed in products that are probably a little bit more cookie-cutter, but a lot of what the London market does is by its very nature not cookie-cutter.
Abbie Day 4:30
Yeah, it's hard to kind of bottle that secret sauce and deploy it with an AI. I think when you move into those more specialist or niche areas, governance obviously plays a huge role. I feel like it might be slightly easier to do effective governance around, say, the very cookie-cutter lines. But when it comes to those specialty or difficult lines, how big of a role does AI governance need to play? And how is Lloyd's facilitating that governance structure?
Rachel Turk 5:00
So from a governance perspective, at Lloyd's we have the same expectations of our managing agents: that their governance is very strong, their risk management et cetera. And that's part of principles-based oversight. Whether they deploy AI or not, we have the same expectations that they have policies, procedures and a clear audit trail of governance decisions.
So if you then take AI and put that into the mix, what do you need? The ones that I've seen that are the easiest are ones where the AI code, for want of a better word, is written in natural language, i.e. English. So then you can trace it through and there's an audit trail. If it's written in binary, one, zero, zero, zero, one, I mean, who knows what that conversation is.
Our reliance is: is the managing agent able to govern the decisions that AI is making or the implementation that they're doing? Have they put the policies and the procedures in place? Do they have the audit trail? Are their second line risk function and their third line internal audit function doing the right things? That's where we're coming from. I don't think at this stage we should be doing, or need to be doing, more than that. We just set the expectations for the managing agents: deploy AI or don't deploy AI, we expect the same from you in terms of governance.
Does it become a bigger issue down the line? I would hope not, because with sensibly managed companies, there's no shortage of information around the risks of AI. So if we're looking at it from a governance lens, a sensible management team is able to see all of those and work through them. That's what we're looking for: the evidence that they're aware of the risks and they're working through them. I don't think we want, from a Lloyd's perspective, to be prescriptive about how they do that. We want to rely on the management team being able to think that through themselves and demonstrate to us that they've put the policies in place.
Abbie Day 6:35
Okay, and is the EU AI Act that's coming in soon going to make any difference to that, do you think?
Rachel Turk 6:41
I would hope not. The whole point of principles-based oversight is that it's about outcomes. It's not prescriptive and rules-based, you must do X, Y, Z. The outcome is that your governance needs to be strong, whether it's governance over AI, governance over risk culture, whatever it may be. That shouldn't make any difference.
So ultimately, principles-based oversight is looking at you as a managing agent: you work out how to demonstrate to us the outcomes that you have achieved, rather than us saying, tick these five things. That goes back to minimum standards. And that's why I think we have to really shift the onus from Lloyd's saying, this is what you need to do, to the managing agents going, this is how we have demonstrated the outcomes of good governance.
Abbie Day 7:26
We've spoken a bit about the efficiency benefit that can come from AI. But I'd like to dig into what the real value creation is with AI for an underwriter. I think it's quite easy to look at technology for innovation's sake. What do you see as the real value creation? Is it the efficiency benefit or the potential to improve decision making and data augmentation? Where does that value sit for you?
Rachel Turk 7:51
I think it's probably all of what you've said. If we take the underwriting life cycle, it is still an inefficient, process-driven process. A submission comes in from a broker and you might not have all of the data that you need. So then you have to go back to the broker and say, actually, it's missing X, Y, Z attachments, could we please have this? And then, oh, I need the loss run, or whatever it may be. So that's quite inefficient.
If AI can see a submission when it comes in and then talk to the broker and go, actually, we're not even going to put this through to the underwriters until we have all of the information that's required, and if the AI is on the other side of the broker, that's seamless. It's not email traffic going backwards and forwards. Then you end up with a fulsome submission, which has taken likely seconds rather than days.
Then you take that through and you go, is this a submission that's sitting within our likely risk appetite? If I'm an underwriter and I have a hundred submissions to work through and I've only got time to work on 20, I want to work on the 20 I'm likely to bind, not the 80 that I'm not. So can I triage those submissions much more quickly? Can AI triage them and go, these are way out of risk appetite, here are the ones you want to work on? That feels like a huge drive for efficiency for the underwriter, so that the human overlay is being used to its best effect.
Then again, simple things. Some of the pilots we've seen fill in the rating model or look at driving better decision making. The classic example is when you get a hundred engineering reports, the best underwriter in the world doesn't open all 100. They open probably the top ten locations, maybe. But there might be something hiding in location number 40 that's actually a problem. If AI can read all of those and flag what the issues are, the underwriter makes better decisions because they've got more data and more information. And therefore your pricing is better, your loss ratios are better et cetera.
And all of this is much quicker. So you respond to the broker more quickly and you have a better chance of binding the deal [check audio]. If you've got the right price and you're quicker, that's pretty helpful. So across the underwriting life cycle, there are various elements. Some is pure efficiency, some is data augmentation and some is just providing much better risk analytics so that the underwriter is more informed and able to make better decisions. That's where I think AI can come through.
If you take it onto the claims side as well: a long time ago now, when I was underwriting, e-discovery costs for big claims were running at around a million a month. Big numbers. And they quickly run through retentions and into primary layers on casualty claims. If you can have AI running across all of those claims files, spotting issues and working out defence strategies et cetera, that has to be more efficient. I think there's a huge amount you could do on claims, in terms of claims efficiency but also better claims outcomes.
I don't think we should just be using AI for the sake of efficiency. I think you're missing a trick if it's just about efficiency. The data augmentation and the ability to make better decisions are really key. Maybe it's an age thing and sort of being a Luddite, but I am still of the view that underwriting is half art, half science, which is the way it's always been described to me. The science piece, AI absolutely can make better: the data augmentation, pulling in different external data sources, whatever it may be. The half-art piece, I think that is still a huge element in this industry. That's human interaction, relationships, knowledge of what your market is doing, knowledge of trends happening elsewhere, knowing what the right price is, which might not be the same as what the technical rating model says but you just know from a lot of experience. I think AI will struggle to replace that. But for a lot of the rest, AI can help you make better decisions much more quickly.
Abbie Day 11:32
I'm interested that you mentioned both the claims and the underwriting side. Is there a way that AI can bring the market out of that siloed way of working, bringing the claims information into underwriting and the other way around?
Rachel Turk 11:45
Without a doubt. Claims really is at the forefront of loss development trends. They see all the loss development trends. And if you're not having those reflected back into your underwriting and your pricing model, you're really missing a trick. You need reserving as well. That lovely triangle of reserving, claims and underwriting data: if it all flows nicely, you have much, much more informed underwriters.
What the London market, Lloyd's and actually I think most insurance companies have struggled with, quite frankly, is matching the claims data and the underwriting data together. Sometimes it's really simple things. Take the example of Alphabet, Google's parent company. In underwriting files it will be known as Alphabet. But in a claims file from ten years ago, it could still be listed as Google. The same with Meta and Facebook. So you're sitting there going, how do I match it up? Because it just doesn't match.
AI can suddenly give you an easier way of matching your losses. We've always struggled with name matching: the company changed its name, but in the claim it's still listed under the original name, while in the underwriting files it's evolved. That's always been a huge problem and AI can fix a lot of that. So suddenly you've got much better flow between claims data and underwriting data, and the ability to spot early loss trends that can then help change pricing models, so your technical rate accuracy is better. That all feels like where the industry should be going and it should be hugely beneficial.
Abbie Day 13:09
That's really interesting, that it's not necessarily a huge technical overhaul. It's quite a simple thing that AI can do that really makes a huge difference.
Rachel Turk 13:17
Yeah. I think with AI, it's not a huge transformation that's required. I think if you can iteratively test and do small tweaks as you go, like name matching, which makes quite a big difference, that's easier to do. And it also proves the use case. Rather than going, I'm going to do a three-year transformation using AI, which just doesn't tend to work, you're saying, we're going to take three months to do this and then move on over the next three months. Then you're much more likely to deploy at scale, much more quickly.
Abbie Day 13:42
Okay. And looking ahead, what does good or great look like for you? What should a market using AI well look like?
Rachel Turk 13:50
Look, I don't think there's a single answer to what good looks like. I think good is up to each individual managing agent and how they want to deploy it: enhancing underwriting insights, speed of underwriting decisions and portfolio management, whilst keeping that human judgement element, enhancing data quality and having robust governance. That to me is good. But again, I come back to this: what good looks like with AI is strengthening human judgement, not replacing it. And that to me is really key.
Abbie Day 14:20
Okay, and it would be great to know what you think are the biggest challenges the market faces with integrating AI going forward. Is there anything it really needs to overcome to reach that good point?
Rachel Turk 14:30
I think part of it is a willingness to try. The challenge I see from a lot of managing agents I speak to is working out what to focus on. It's a classic prioritisation problem. As I said, there's no shortage of ideas, but you've got to focus on a few things.
The other challenge for the market, honestly, is that we've spent this time talking about AI deployment at the managing agents. They also all have to think about AI deployment by the end customer, how that changes the underlying risk profile and the AI liability that is going to be around the market, and what they are going to do with that. From an underwriting lens, the underwriting of AI liability is a totally different topic from how AI is going to be deployed at the managing agents to improve efficiency.
Abbie Day 15:15
And it's definitely a topic we're going to cover on the podcast, so stay tuned. Rachel, thank you so much. It's been absolutely fascinating talking to you about AI and the London market. Thank you for joining us.
Rachel Turk
No worries, thank you very much.