Tech-Driven Transformation: Google’s Take on Insurance Evolution | Nigel Walsh, Google

by Juan de Castro, COO Cytora

Juan de Castro: Welcome to another episode of Making Risk Flow, and today I'm joined by Nigel Walsh. He's the Managing Director of Insurance at Google. I think you're one of these people that nobody knows how you do everything that you do. Actually, I've got a very strong hypothesis that you are one of these AGI, Artificial General Intelligence, agents, because I think the chances of you being human and being able to do everything you do are quite slim. If I look at your background, you've advised insurers, I think you were a partner at Deloitte Digital and other consulting firms, you've advised startups, you were a mentor in a startup bootcamp a few years ago, you've even written books. I think probably what you're missing is to plant a tree, right? Is that the one?

Nigel Walsh: I’ve actually done that as well.

Juan de Castro: So you've done everything then. And I think for many years, you also led a podcast, and you're quite involved in other industry podcasts. As always, Nigel, it's a pleasure to have you join us and catch up. Thank you for joining us. 

Nigel Walsh: Thank you very much. I'm excited to be here. I'm a fan of the show. Actually, I've listened to a lot of podcasts that people on the teams across Google should listen to and you're in the list as well. So it's a strong recommendation from me.

Juan de Castro: That's very nice to hear, thank you. Let's start with a brief overview of what you do at Google now.

Nigel Walsh: I lead insurance for Google Cloud specifically, and that is in essence helping insurers, brokers, agents, and insurtechs get the best out of Google Cloud's capability. It's funny, I was at a career fair last night, and people were asking exactly the same question. Hopefully, I've got my answer down to a T. Imagine all of the wonderful technology that we've got inside Google to help people be more efficient, find information faster, take out some of the busy work that we do and make that readily available to every carrier, broker, or agent that's out there. That's, in essence, what I do. We have phenomenal technologists across the business that are in development, engineering, and so much more. I always try to take a business angle to say, actually, what does this mean for an underwriter and a claims adjuster, somebody in pricing? I try to join the dots between the wonderful technology and the so-what for what it fixes on the business side.

Juan de Castro: Which is great. I think this is what more and more insurtechs and technology providers should do. It's how we use technology to solve business problems and help, to some extent, educate the insurance industry on how technology solves business challenges.

Nigel Walsh: I am a massive technology nerd, I have been from a very young age, but it can't be technology for the sake of technology. It has to be useful to the thing that we're trying to do in our business. Obviously, every industry is different. Every insurance carrier is at a different stage in their journey in terms of technical debt, maturity, adoption of AI, and all those good things. For me, it's helping our teams across the world. There are hundreds of people who focus on insurance across Google Cloud globally, how do we get them to say, if your customers are this level of maturity then these are things that we should be going after if they’re at this level of maturity then these are the things that we should help them or how we can get them to leverage Google Clouds Technology. 

Juan de Castro: That makes sense. And as you said, you're working with many insurers now and in your previous lives. What do you see as the major shifts in terms of what specifically insurers are focusing on right now from a technology perspective, and also from what business problems are they trying to solve?

Nigel Walsh: It's interesting. The last 12 months have really shifted back to, hey, how do we get the most out of our business? So I'm fortunate enough to have a global purview over what's going on across North America, Asia, in Europe. I've always said, to me, Europe is very mature in terms of adoption of technology and capability and probably further ahead than many other territories. I would say Asia is equally similar. The US has just got a ginormous scale in terms of the things and the number of people that we engage with, whether it's agents or brokers or carriers themselves. I think there are 6,000 insurance organisations in North America. The focus for the last 12 months or so, though, has really truly been, oh, my God, our combined operating ratio has just gone through the roof. How do we take costs out? Because of the admitted versus not admitted lines business in North America, to correct that versus something in Europe or elsewhere takes a little bit longer. I think one of the CEOs came out and said, “for us to adjust and get right back into the market, it's going to take 12 to 24 months to solve that”. So there's been very much focus on the, whilst innovation is really important, it will always be important, it is actually, how do we take costs out? How do we optimise our processes, whether it's on the submission process and commercial lines, whether it's on how we buy and how we engage, or what we do in contact centres and stuff like that.

Juan de Castro: Well, is this part of the evolution from a hard market to a soft market? I mean, it's almost like the tide goes down and you start realising you just stop making that much money. Or is this driven by that or something else?

Nigel Walsh: I think it is. We're in a very fortunate space. People can say insurance is recession-free in that way. People will always need to buy insurance. There's definitely a lack of capacity in certain lines of business, that's how it's causing prices to go up. You look at the news in the UK, you see many carriers pull out of the EV space. Now, the EV space, I think, I'm a massive petrolhead, I love cars, I've got an EV at home, love talking about them, but if you're making an unavoidable link back to the energy crisis in Europe and all the other external factors that we have no control over, there's lots of press in the UK talking about rate of insurance costs going from £600 to £4,500. That makes buying insurance and actually the transition to cleaner energy really, really difficult. I don't see the same challenges in the US because they don't have the same energy crisis out there. So actually, there are different factors affecting different territories. I don't think any insurer I've spoken to in the last three years whilst at Google has said to me, we can't be more efficient. Everyone is striving for efficiency. And even in the last 10 years, I'd say across my career in Deloitte or Capgemini and now here, we've always been after the same things. I often call it evolution and revolution. Evolution, how do we go after the 2%, 5%, 10%? Or actually, what's the revolution? How do we go in there and say, if we were going to start from scratch, what would it look like? And I think everyone would agree it probably wouldn't look like this.

Juan de Castro: Agreed. And almost when you're trying to get expenses out, obviously, I mean, you can do the quick solution of just getting rid of people, which is completely unsustainable.

Nigel Walsh: And lots of that's happened. There's been a huge number of layoffs across the insurance space, that are thousands of people in the frontline, thousands of people in the contact centre, where they have reduced staff to take costs out immediately.

Juan de Castro: Exactly. Which is a short-term action. Ultimately, you need to find a better way of operating to make that sustainable, right? Or it's a much more sustainable way of reducing your expense ratio if you find better ways of working. So is that how insurers are thinking about it now? Is that about the operating model and the writing workflows, et cetera?

Nigel Walsh: You still need people. Insurance will always be a people business no matter how we engage. We're not here to remove everyone from the entire process. We're here to help those people become 10x better, not just 10% better. So I still think, whilst you're seeing some attrition going on across insurance in terms of people and stuff like that, there was an article last week in one of the insurance press that talked about 400,000 jobs to go in insurance across North America, if you dig into the headline, it's natural attrition. It's not through AI or anything else. It's natural attrition. We all have an ageing workforce. Wherever you look inside insurance, people will retire, and they retire at different ages. And we can get much more efficient with the processes that we've got. If you look at the worlds that you focus on and that I focus on, the steps that we go to to get a submission through the door and the busy work that goes on is not sustainable. It hasn't been sustainable for a very long time. And over the years, we've been digitising it bit by bit by bit, but it definitely needs to go through a revolution.

Juan de Castro: I love your idea of busy work, and I would love to touch on that a bit more, but is that revolution then? The insurance industry has been talking about digitisation of the industry for 10, 15, 20 years. Is this revolution about almost rethinking the underwriting workflows? Just rethinking how processing risk works, which in the end is a risk that goes from a client to a broker, to an insurer, to a reinsurer. How does that workflow look in a way that is 10 times better? 

Nigel Walsh: There’s a couple of ways I look at it. If I was looking at anything to do with digitisation, which I think is hugely important, it probably falls into the evolution camp because we're taking an existing process, we can walk outside the street here, walk down to Lime Street, around Lloyd's, and we still see our folks walking around with big packs of slips and whatever else. That will be digitised over time. And you and I are gentlemen of certain ages. Our kids, our next generation, I suspect, won't be doing what we did or what our parents and grandparents did. So that will all be digitised without a shadow of a doubt over the coming years. I would also say things like COVID, whilst a terrible thing that the planet went through, was a huge accelerant for digitisation across the London market and insurance more broadly. There was no choice. So we had this huge push to digital. And now we need to shrink some of those processes down to make them much more efficient. We take out steps of those. The revolution part would be, and we talk about this all the time in different steps, but what if we were to take external data, your internal corpus of data, all the history of stuff that you've got inside your lost funds and your previous history of working with clients, whatever else, and your appetite, and all this new capability that we've got and started to make much faster decisions before you even got to a human to review or presented them with the things that they wanted. I think that workflow could be shrunk completely from multiple steps to one or two to give the underwriter the ability to do the job they were hired to do in the first place, not to do the busy work of interacting with 5, 10, 20, 30 systems and paper, whatever else, so we could actually engage in the, hey, I'm a risk manager, I can understand what you're up to, let me tell you about the things that are important and things that are not.

Juan de Castro: In the end, it's this concept of how technology and data can do the pre-analysis of a risk. At some point, they will be better than humans at doing that pre-analysis, right?

Nigel Walsh: But it's all good. It's very much how we sit side by side to augment you going forward. And I think over time, you'll see lots of risks, in essence, become commoditized. And when I say commoditized, you've already seen what folks like Ki are doing on the follow side of the market. Before, that was all manual and whatever else. Today, you've now got algorithmically underwritten follow syndicates. Ki was first. Apollo is entering into the market. Many others will follow into that space. Maybe you'll see in five or 10 years' time that the follow space is entirely driven by algorithms. We've seen it in capital markets and elsewhere. There's no reason why that couldn't be done.

Juan de Castro: I remember talking to, you might know him, Roman Itskovich, one of the At-Bay co-founders in the US. And he described a beautiful framework. I think that summarises what you're saying, which is the evolution from human in the loop to human on the loop to human off the loop. I think it's a great framework to describe what you're saying, which is that the insurance industry traditionally was a human in the loop. Every step in processing a risk requires human intervention, like a human actually clicking on a screen, and doing something about it. From extracting data points from sleep to getting external data to inputting all the data fields into a pricing engine, etc. That's human in the loop. I think what you're saying is now there's an opportunity to move to human on the loop. But technology is doing all those busy activities, pulling all that data, doing the pre-analysis, and generating all the insights for underwriters. And you just want to deploy that underwriting or human capacity to do a final review of that analysis that technology has facilitated.

Nigel Walsh: Completely. Think of yourself as an airline pilot. An airline pilot has all the data of all the things that are going outside, all the things that might happen in the future like weather conditions or future storms and stuff like that, as well as all the things that are going on with the aircraft. They've brought together a dashboard that allows you to make decisions going forward. The in the loop, off the loop description that you've given, which I love, is actually like insurance. If you go back through the history of insurance, insurance was there hundreds of years ago after the event. Today, I would say it's with you at the event. And in the future, how often have we talked about going from reactive to preventative? So imagine if insurance in the future is all about predicting and preventing, and actually getting to the point where you actually never need to use insurance in that way. So it's very much following the same sequence.

Juan de Castro: Yeah. I almost like the next step will be just building on that framework of human in the loop to on the loop. The off-the-loop is that once you digitise the process, I mean, you can be quite surgical about what type of risks require human intervention versus what risks you're happy to put straight through process. But then you need that first digital workflow to be able to start making those decisions.

Nigel Walsh: A thousand percent. Without that agility, we can't write new business. Without the capability, we're still stuck in Google as a technical debt, legacy applications, and so much more. We're not able to bring that dashboard to our underwriters, our underwriting assistants, or our business portfolio to go, actually, we've got too much concentration in property over here in North America, let's pause on that and do these other things. We need to be able to make those real-time decisions in a much faster, more intuitive way than we've done previously.

Juan de Castro: Without getting too technical, you're building on the point you made a few minutes ago about getting all the data insurers already have, then combining that with the data about the risk that the broker is providing, combining that with external data sources that give you more information about the risk. Then is your vision that with some of these AI technologies, we can combine that, probably also pulling things like underwriting guidelines and loss experience from previous similar risks, et cetera, and with that, providing the underwriters not just information that the broker has provided, but the really proper insights about, okay, this is the risk you need to analyse. These are some of the comparisons with similar risks you've seen before. This is what we've seen from a loss experience perspective, it's happened to similar risks. That is the type of insight an underwriter needs. 

Nigel Walsh: Yeah. My goal here is to see how you get the connected dots between all of the underwriters, respective of class or line of business, to say, imagine having the ability to have your best underwriter, your best-experienced person on your shoulder at all times and have access to their knowledge, of which in many cases is 30 years of history of writing this class of business people. They're experts in their field. How do I get that at my fingertips all the time to say-, lots of the carriers do this today, they've always got external data, past history and lost runs. New data sources can be compiled with third-party sources to make sure we've got accuracy and so much more. But imagine we're writing a satellite. We've got information that other people have done in different classes that we don't know, but that's now at our fingertips. And we've now got the ability to go, actually, did you know in this line of business with this class, here's our experience, and this is what might go wrong or what might go right. How do we help you ask better questions to the person looking for insurance in the first place? So that's got to be critical to make sure that we have that best individual looking at the risk at all times.

Juan de Castro: Yeah. And what is different now from five years ago? Conceptually, this vision is probably not that different from the one the industry had five years ago, but there have been major changes in technology. So is it the difference that now it is real?

Nigel Walsh: Yeah. I remember I started my career at an office furniture manufacturer, and I was in technology. And back then, this is going back 20 plus years, people would go to get information from the invoice room or the purchase order room or whatever else, and there were boxes and boxes of paper where you'd go and find and file the information and go pull it out and go, okay, I can now close this case out because I've gone and got the stuff. Over the years, that's been digitised and so much more. What we're seeing now, though, I think what's different between 20 plus years ago and five years ago, is our ability to understand and dig into that corpus of information. A, it's been all digitised, and B, our ability to understand and bring you the pieces that are irrelevant at any one point in time has been tuned massively, i.e., the evolution of large language models over the last 12 months more specifically, we've been doing AI. One of the carriers came out and said in their interviews a few months back, we've been doing AI for six or seven years. And that's absolutely true. Let's not knock insurance here. They've done a really good job. There are pockets of brilliance across the industry where folks have made big inroads to do things that are different. What's changed now is that technology keeps advancing. None of the folks like Google or elsewhere have slowed down in the development or evolution of these technologies. And now the ability to give something like Vertex AI, Google's document enterprise perspective, along with our modern garments to take large language models combined with private sources of data, adapt the layers, and so much more to actually combine all these things in a really efficient way and give that to the business rather than make it a technology-only thing. Now we're putting the capability in the hands of individuals. A great example is when I sat down with a friend of mine for coffee a few weeks back. He's a lawyer in the city in property. And he was asking about, what is this stuff? Will it ever take off? And what does it actually do? So I opened up Bard, just as an example, Bard’s our consumer version of a large language model. I've put into that a link to an insured short-haul tendency, so a 40-page PDF for rental contracts in the UK. And I said, ask me a question. He went, go away. I said, ask me a question. And I made a 40-page PDF conversational with legal questions in a heartbeat. And he's like, this is unbelievable. We've also then taken it to a salesperson and said, imagine if we took a 1,500-page Basel II compliance document and said, is X contract compliant against these conditions in the Basel II contracts? So our ability to pass at pace and speed at a business user level, not technical, to be able to put those things into the hands of business users has dramatically changed. What I was doing in Google 18 months ago took me six or eight months, I'm doing today in Google in four to six weeks. So the speed of change and the ability and the ubiquitous nature which everyone's trying to devour has, in my mind, changed the world dramatically.

Juan de Castro: And I think building on that, if I had to pinpoint to, which I think you've touched on, what is the single thing that is significantly different now from the AI three years ago is the ability to use those models without training. In the end, I think your experience from this coffee place with your lawyer friend is, I think what amazes your friend is not just that the model answers the question, it’s because there was no training required. It was literally, you just uploaded the PDF and the model was able to answer your question. And I think that is what we're seeing with many of the insurers now is they said, well, yeah, six months ago, we had to train these models with 5,000 examples for six months. That would be quite a huge cost. Also, the time to value was very long. Whereas now we can get similar outcomes with zero shot, but in real time on this, right?

Nigel Walsh: We look at some of these offers that we're working with right now. We'll see a thousand inputs into this or a thousand examples. There's a really good way to start training some of these things. So it's actually either zero shot or a few shot learning in this instance. Number one, I should also say the idea that one model will solve every problem in the world is probably the wrong one. We've always said, actually, there's not one model to answer everything, which is why our approach is within the Vertex AI platform or Google Cloud, we've got this capability called the model garden. The model garden has support for our core foundational models, whether it's text, image, chat, voice, or whatever else it might be, but then support for other models that are out there that are open source as well. So whether you want anthropic, metal, or whatever else, you can use those inside Google Cloud with the same security permissions and everything else that you've got. So that's really important. But as you go through these things, what we want to give you the ability to do is tune and fine-tune those specifically because one of my colleagues always says, “You don't necessarily need to know the back catalogue of Taylor Swift, but we do want a model that's focused on commercial property or legal conditions in property underwriting across North America. Could that be a composite data that you use specifically against things?” So it's not going to be one answer. It's going to be, how do we combine that, and that, and that to solve the problem?

Juan de Castro: I think my takeaway from this part of the conversation is clear, new technology does accelerate time to value to solve some of these problems.

Nigel Walsh: So I will add, it always will. There was a great quote by Justin Trudeau who used it. I think it was his quote. He has said something like, “Technology will never move this fast, but never again will it be this slow”. And I love that because it just says, we're just going to get faster and faster. The other thing I would say that I think has been really unique is there's been loads of trends. You and I have seen it all over the years, whether it's blockchain or whatever else it might be. I've never seen in my career the adoption by boards where they said, “What is generative AI and how does it affect my business?” over the last six months or so. Six months ago, people would say to me, “What is it?” Three months ago, they were saying, “Well, how do we use it?” And today they're going, “Oh, my God, we have so many use cases. Where do we start?” But the adoption has come, the interest has come right from the very top of the organisations that have said, “We'd like to get this”. And I think the other example is my son at 14 has been taught how to use this at school. So I don't have to worry about the next generation of the workforce understanding whatever else. It's already there. And they're using it in their everyday lives. They've been told how to use it in school, and how not to use it in school. Don't get AI to write your essays, kids. That's the wrong answer. But get AI to help you with your research. What do we do as entrepreneurs and claimants just like everybody else? We do lots of research. So I don't worry about the next generation. And I love the ubiquitous nature in which every CEO, every exec suite's gone, how do we apply this to our organisation fast?

Juan de Castro: So at that point, you have that conversation with the boards or with the executives. What's happening next? So what are insurers doing about it now?

Nigel Walsh: There will be some great studies out there from McKinsey, Deloitte, BCG, and others to say, how has this been adopted? What are the fears? I walk into most board meetings and the same questions arise every single time. How do we cope with security? How do we cope with IP? What people do we need? And those are broadly the big questions that people worry about. On the people side, do we need to go hire 1,000 data scientists? Well, hopefully, these examples we just talked about in terms of its ubiquitous nature, the fact that I can take a pull-up at a coffee shop and show you what the consumer tools can do. Imagine what the enterprise tools can do. So it's not 1,000 data scientists. It's probably a mix of a good underwriter or good claims adjudicator, along with a technology person who will do something in a sprint very, very quickly. So how do we redesign our organisations to have those right teams that can move at pace? On the security and IP side, early on, I would say there was a good mix of organisations that said, let's ban access to everyone because we're not sure what's going to happen or let's give everyone access to work at how we get an understanding. Over the last six months, I would say it's gone from 50-50 to 80-20 or even 90-10. So the bans on using it have almost dropped entirely and organisations have created safe spaces and the capability for their employees to try to use it, and deploy it internally. And I've often talked about, obviously, we're a risk business ultimately, so how do we deploy this in a crawl, walk, run phase that says let's use it internally first and foremost. It might be an internal IT or help desk case or HR case, moving into conversational stuff that's not using third-party data or your own data, so it might be smart chatbots that can help you or do the triage at the outset. And the run phase might be, let's combine this with your data, our data, and everything else to actually give you the risk appetite or claims insights back to those folks as you go forward. The big question for me is how do we get started? And that's where I think lots of organisations are still mulling over the right approach and putting the right governance and teams in place to move fast. The stats are out there still that say most models that people look at don't get to production and we haven't got the right organisation, governance, and setup in place to make this successful.

Juan de Castro: What needs to change there? Does it start with a strong vision from the board or the executive team? So how are you seeing that?

Nigel Walsh: The desire to get this done is without doubt, you've probably seen the same thing, everybody wants to do something. How do we give our teams the authority to get started? I've always had a mantra, think big, start small, act quickly. I'm not sure insurance has always followed that. They've always been great at vision. Insurance has never been described as fast.

Juan de Castro: It's execution.

Nigel Walsh: It's key. And I think what we've seen in the ones that are successful for the folks that have engaged already is folks that have made a fast start, built confidence, and iterated over time. Some of the use cases that we try aren't going to work.

Juan de Castro: And you touched on one of the things there at the beginning in terms of what hurdles to overcome to even start going was IP. I want to touch on that quickly because I think that is something many of the insurers I talk to are still concerned about. It's like a gate. If I'm sending my data to Google Cloud, it's like, what they will use it for, et cetera. I'd love to hear your thoughts on how you overcame it from the Google perspective. And probably there's an opportunity to better explain all the safeguards you've got and others have.

Nigel Walsh: Yeah, let me give you two examples. So example one, how the platform has been structured, the generative AI stat, is designed around bringing a copy of the model to your instance of cloud, and ensuring that whatever is in your data stays yours, and none of that gets passed back up to the model, that's frozen and vice versa the other way around. You will never be able to download that or whatever else, but bringing those two things together, the output of which is stored with yourself. Number one. So that's important. Actually, back to my point about security and IP and all those good things, we didn't rush out with our answer to this. We had a very thoughtful approach to allow enterprises to build with that in mind and security and everything else, which is key. Number two, we've also published, I think it's July timeframe, maybe a little bit later, how Google is going to indemnify organisations and actually the work that we do in two-fold. Number one, the training data that we've got, we will make sure that we indemnify anyone from saying that if you're allowed to use it, you're okay with that going forward, number one. And number two, the output of the generative models will also protect you against, provided that you didn't maliciously go in there and say, please go and copy something that you shouldn't have tried to copy. So we put those safeguards in place to give organisations confidence that actually we're there to support you, both on training data that's been used correctly and the outputs from generous opponents, which I think are key to giving organisations confidence that we're here from an enterprise perspective rather than anything else.

Juan de Castro: I think there's still an opportunity, and this is more advice of almost like creating one slide that captures that because I think that is a question that comes up in pretty much every meeting we have with insurers.

Nigel Walsh: Yeah, I agree. I think it's fueled by all the good stuff that never gets the press headlines, but all the bad stuff always gets the headlines and says someone's accidentally uploaded something to a third party model, and now all of that's been used to train the model. That just does not happen in the world of Google.

Juan de Castro: Yeah then moving on to how to get started? Especially in these environments where technology is moving quite fast. Obviously, we didn't have LLMs 18 months ago and now we have them, et cetera. Would your recommendation for insurers be to start by deploying an architecture that allows them to communicate with these latest models? So almost creating the basic infrastructure allows them to use those external models and the output for underwriters, and then be able to plug and play different models as technology evolves. Is that foundational work the first step? 

Nigel Walsh: Spot on. Number one, absolutely spot on. None of this is any good without any data. So we've got to have access to the data, first and foremost. You've got to have the environment set up. Our goal very much is to be foundational in all of these things that you do. So the work that we do with Cytora, for example, allows you to take the Google Cloud platform and have connectivity to all of the models, both our own foundational ones and third-party models. But you've got access to Vertex AI at the outset. You have the ability to deploy those models very, very quickly in a secure, controlled way. And the things that you were doing 12 months ago or 18 months ago that have been augmented by this new capability is thereby equal. And all the things that we're releasing over time and the access that you've got to Palm 2 and so much more will keep coming as we release these things into the environment. But what's critical is that you have that core foundation that allows you to actually go test and try these things and over time release those to different parts of your participants as you see those get more comfortable. So data is critical in all of that.

Juan de Castro: Also, I think one of the things you do really well at Google is this true partnership with players like ourselves, like Cytora, in terms of really building this together. I think this is something that I want to compliment you on because I think, at least from our perspective, that's working really well.

Nigel Walsh: That's a huge thanks down to the staffers, Katrina, and the rest of the team. So yeah, without question, it's great to hear.

Juan de Castro: Huge kudos to both of them and others. So then zooming out again, you talked about some of the concerns, how some of the insurers can start going. You're looking at the whole market now. And perhaps this might be different in the US and in Europe. Do you see most insurers are making progress? Or are you starting to see a two-speed?

Nigel Walsh: It's funny, actually the words I used the other day were, I do find that insurers over the next 12 months will fall into two categories. Those that have and those that have not. And almost every carrier that we've engaged with partners like yourself are into the category of have. And what they'll be interested in now is how quickly they can deploy that across their organisation for the right use cases. I still feel there are too many organisations that are still thinking and pondering about how and worrying about, rightly so, the risks, but not putting the right guardrails in place to go, well, how do we test it with confidence in a secure, safe environment? That gap will create a fundamental difference between organisations that have and have not. So, in your words, that two-speed use of AI across the organisation will be really interesting. Fast forward 6, 12, 18, 24 months. I think we're at the very tip of the iceberg in terms of the use cases that we've got. We've almost got an accelerated automation, augmentation-type use case today. Once people become confident and capable with the technology, they've got a better understanding of it, and they've got that foundation deployed, I think the use cases are going to get really, really interesting. I was speaking to the Society of Actuaries a few weeks back in North America, a really interesting opportunity for what they're going to do and how they're going to work with their model skateboard. How do we take all that formative information and that data that we've got now and start to design better products based on the things that we've got that we could never even dream of analysing previously, but we've now got the ability to just go query in different ways and say, imagine we could do this, this, and this. What does our product need to look like that we should design and take to market? Or imagine asking models, if I want to improve my combined by X percent, what things should I do or shouldn't I do? Imagine getting that guidance. Back to my airline pilot that said actually your risk appetite is too high or too low or these are future events that will go on. This is what you need to change going forward. So we're going to allow all those people to make much more real time decisions with all this information in their hands, 

Juan de Castro: And that these distinctions of those who have and those who haven’t, I think I do see our responsibility as providers for the insurance industry to also help out that who probably haven’t because, in every industry, there will always be early adopters and late adopters, right? And, often, the late adopters, have not yet joined the transformation, mostly because of a lack of understanding or because they have not been exposed enough. And I think if I had to say what is the main thing we can do to help those late adopters or those who haven't group is almost, I think this is a responsibility for us to stop talking about AI and technology and talk about what is the business impact of what the other group is achieving. So the other group is achieving 10 times faster broker turnaround times. The other group is achieving twice or three times more productivity per underwriter. And this is how it's achieved. I think that the late adopter group needs more business language and potentially less AI. Would you agree? 

Nigel Walsh: I’d agree. I was at a board meeting recently where one of the CEOs described this, he goes, “Nigel, I'm confused. Is this traditional AI, old AI, or new AI?” And I think everything's mixed up quite often. I'm a great fan of show not tell. So can we demonstrate this to you with some of your information in a safe environment very, very quickly? And it doesn't have to be on the broker side or the underwriter side. It could be, as I said, an HR case or an IT help desk case or something that's internal to them that we can start to show the power. Like my friend I showed the legal contracts to, it was an eye-opening moment to imagine it. I'll give you a great example. I was at home at the weekend. I'm not a technical person, as I said. I got an error message on one of the systems that I've tried to make work with some connected home stuff. I had no idea what it meant. I picked up these multiple lines of code. I dropped it into Bard and said, what does this actually mean? And it explained it bit by bit by bit. And then I said, well, how do I fix it to fix that piece? And it broke it down in layman's terms for me the things I needed to do and install on the Mac to make it work in my club environment. And having no experience of code, no experience of that language, I as a layman could actually take that, better than going through any manual or elsewhere, it was very specific. So those sorts of use cases when you go, oh, it's now told me the answer to the question. The other thing that I love about what we've done at Google is the citation links. And that helps build confidence, not just, here's the answer. I've often said that AI is like my 14-year-old son. He'll give you a really confident wrong answer. So what we want to do is give you the link back to where it came from in the first place. So I have this answer to my question with a link back to where it came from in the first place to go, oh, I believe that. Now imagine that in an underwriting perspective where I've gone, here's my risk. Here are all the conditions that have been met. Here's my summary of what I've told it to do and the table I've told it to, with links back to, well, this is the last word it came from, this is the Accord form it came from, this is the email it came from. Now I've got everything in one place with the ability to go back to each and every one of those sources in the first place. So you've built up confidence by giving them the evidence of where it's come from. So it's almost back to the other things we get all the time, Responsible AI and Explainable AI and all those things. These are all things that we've built in from the outset to give you that confidence from day one.

Juan de Castro: I love it. I fully agree. Almost like that is a very nice summary of the whole conversation. How do we give confidence to the users? How do we help accelerate the early adopters? How do we get the late adopters to not fall behind?

Nigel Walsh: My biggest request is do nothing is not an option, and I think there's very few that are doing nothing. Even the late adopters that are pausing have a consistent approach to working out where they go. It's just that doing nothing is not an option for a number of reasons. A, from the business point of view, first and foremost. But even I hear this all the time about how we go and attract talent and everything else to the insurance industry. As I said, my son's learning this in school right now. They want to know when they come into the organisations that we are using best-in-class technology, whether it's Google or something else, that we are doing cool things around AI and machine learning and so much more. So I think demonstrating that we have got partnerships with Cytora or Google or anything else is actually a really good thing to say, actually, we're working with the world's best tools that are out there today in our business practices, whether they're Intel or customer-based.

Juan de Castro: Fantastic. I'm going to leave it there. Doing nothing is not an option. I love that. That's almost like the key takeaway. Nigel, thank you so much for joining me. It's been a pleasure.

Nigel Walsh: Thank you so much.