Global Risk Summit AI utilisation, and how to guard against inherent risks

Well, good morning everyone. Um, AI has moved from the innovation lab to the boardroom and the front line of how organizations compete, serve customers and manage risk. This is why the timing of this discussion today is very important. We are well past the point when AI was a technology experiment. It is now a board level issue touching on productivity, resilience, regulation, um workforce strategy and competitive advantage. The question is no longer whether organizations should use AI. It is about whether uh organizations can move from scattered experiments into scaled governed and measurable adoption of AI across the organization and whether they can do so in a way that improves business outcomes and does not create new unmanaged risks. We have a great panel here today with us to discuss this and many other topics. Um, we've got David Griffiths, the newly appointed group head of artificial intelligence at City Bank, bringing us the perspective of scaling AI within one of the world's largest financial institutions. Vishal Maria, founder and CEO of Quantexa, bringing perspective of how trusted data context and decision intelligence foundations can support scaled use of AI and Praa Ladva group chief digital and technology officer at Swiss Reing us the perspective of how you scale AI within one of the large the world's largest reinsurers. So let's get started with the uh discussion. Vishau, if I may start with you, beyond beyond the headlines, where are organizations today in terms of deploying AI and what do you see as the most common pitfalls as organizations try to scale AI throughout the organization? >> Thank you, Max. And um thank you for being here this morning. Um so the world of AI um what we're experiencing today has been a journey that's sort of catapulted in the last three years. So just to take a we step back AI is not a net new thing for things around predictive AI has been here for a number of decades. Um so things around deep learning, machine learning, predictive models, those are areas of the dimensions of AI that's been um in sort of mainstream for a number of years. In the last three years, obviously there's been a big um transformation when it comes down to generative AI. So the ability of large language models, small language models um being deployed at scale um within organizations. Now as we look at this journey in 22 23 I think many of us would have heard things around using AI to create poems. If we go back a couple of years back, that was one of the things creating poems and we all thought that was rather amusing. But what is the actual effectiveness and realism around this technology? But over the last sort of 18 to 24 months and in particular the last 12, this journey is now becoming a lot more of a reality as organization grapple with this technology. This technology is a once- in a generation technology set and as organizations are deploying this enterprisewide either internally within their organizations to help them be more efficient, more effective, if that's in engineering, if that's in how they go to market and service their clients. These have been great examples that we have seen and supported our clients today. But in addition to those internal use cases, we are also seeing AI being deployed right at the customer front office where organizations are using this technology to have better go-to market access but also creating new products with their engineering team to better support their clients. So that's been the journey and we're in three years of that. Right? So if we look at the the cycle of over a 10-year period, you know, some people say, well, are we at the beginning, are we at the middle, or are we at the end? Well, we're in year three and this is a transformational technology set. Your second part to the question is the pitfalls. So this technology as mentioned is transformative, but it fundamentally comes down to a key thing that we see in boardrooms, which is around being on the right side of history. So if we unpack being on the right side of history, one of the key elements is the ability to trust data, explainability, transparency of the data that you are using within your enterprise to feed these models is now a board level conversation. And just to put one more spin and I'm sure my my guest will add to this. These transformations will be successful or not if there is a clear alignment between the board and the teams that are deploying. So we're seeing more where engineering operations and the board objectives are coming a lot more closer together because if you do not have the board alignment these will end up being proof of concepts and not going enterprisewide in organizations. >> So a transformational once in a generation opportunity to transform with important arcs around trust and alignment between the boards and the front line. David, if I can bring you into this, bring us a perspective of how that resonates from a City Bank perspective. >> Yeah, I mean I think um Vish is exactly right and I think AI has been around for a very long time. Um but there are two significant events in the past three years. One, the introduction of chat GPT. So we all became aware of this new technology and then in the last three to four months the emergence of the current crop of frontier models from companies like anthropic which are incredibly capable. Um, and why it's so interesting is the introduction of the models in recent history is what's really seen the revenue start to flow into a lot of these foundational model companies. So last year there was a lot of discussion on an AI bubble. Is it real? I think what we're starting to see now is there is revenue um flowing into the ecosystem. Um so I think from our perspective we've um absolutely um been on a journey and I think to to Vic's point the the need to be very crossunctional and deliberate and intentional about how you approach this is why we put actually our new organization together. So we've been spent a lot of time on some very scalable globally consistent infrastructure. We spent a lot of time working out how to wire and plum that into the bank. And actually now what we've recognized is you need to do those two things together. So we brought an technology mandate and an operations mandate with very very clear support from our board and CEO in terms of what our objectives are and now we're just executing the machine. Um so yeah we're we're certainly beyond the point of you know is AI real? Is it a flash in the pan? It is going to change the way that we we work and do business. >> Pra I'd like to go to you next. You've talked a lot about the AI paradox where we see a huge amount of experimentation and investment into the AI topic by various organizations but not many of them yet seeing the appropriate returns on investment. Why is that happening and from your perspective what is good look like when AI is deeply embedded within critical workflows within insurance and reinsurance? >> Sure. No, great question. And I think I could really simplify it and say it's it's because it's difficult, but it requires purposeful focus and attention. So, you know, you can all read you all look we all looked at the announcements on LinkedIn, you all read press articles and people will genuinely say, I've got 300 use cases, I've got 500 use cases. That is good because from that we learn. We've got to remember this is still a very naent technology you know in the current guys three years and it's evolving every day you wake up and it can do something different. So these proof of concepts are critically important from a learning perspective. But we now need to move to the next stage and the next stage is not thinking about it as a technology challenge but thinking about how can this technology drive real business value. How can this technology help solve industry problems and challenges that we have faced for many years? So, it's flipping the the dynamic on its head and that's why this is a CEO topic. This is a board topic actually first and foremost it's a business topic. So the way you go about the way how we've approached it is purpose very purposeful um very structured and taken some of our core insurance and reinsurance processes and fundamentally reimagined them. So we learned a lot from a proof of concept and then we said let's take a process end to end and let's ignore how we've done it for the last 20 25 years and think about how we could do it in today's paradigm and in a room of 15 people there are probably only two technologists. The rest of them were business folks you know explaining how we've tried to get sense out of unstructured data for years and failed. and and then it's taking those challenges and then using the technology to solve it. So to give you a real example, um engineering and construction underwriting, you know, very data intensive can have to be price competitive and your offer has to get out there very quickly. It can take anything up to 45 50 days to get an offer out there. With the implementation that we've put into our one of our product lines, we can now get that offered out in one day. And that's because the the ag the agents we've built we've built 14 agents that talk to each other and what we've done is then reduce the passing of data from potentially 14 systems to five. Now having said that what's really important is security governance and control. So this is augmenting our underwriters and at every step of the way the insights get checked balanced before they move to the next stage. So it's moving from let's think about a tech issue to really think about as a business challenge and solve some of the things that we've been doing and had have had found challenging for so many years. >> 45 days to one 14 systems to five and a reimagining of business processes. >> It's very impressive. Can I just follow up with a with a question around so how do you decide as you go through that portfolio approach which initiatives to stop and which ones to double down on? >> Yeah. Um and this is when you have to go to old school business cases >> because sometimes when we think about innovation we think about the pace you can't lose sight of what is genuinely makes sense. So all of these things are driven by robust B business cases both from a cost perspective but also from a return perspective. Um but then you think about so that's it's a very like I mean we all run business case processes. So that's the first thing I would say. The second thing I would say is having a balance in a portfolio. What are the quick wins and I call them the speedboats. What are the speedboats that you can launch that can have immediate impact and what are the other things that you need to do that are genuinely going to take time and not materialize for two to three years. So what this also requires is a degree of patience. Um because a lot of people go right we've got AI now what's going to change and you're like actually some things will work some things won't some things will give a return in a year some might take five. So I think the word you use is portfolio and I think that's the way you have to approach it not forgetting how we manage our business and have always managed our businesses through robust business cases. >> So building on that portfolio discipline V like to go to you next. So as you as as we think about initiatives that progress through that portfolio from your perspective, what are the right foundations from a trusted data context decision intelligence that need to be there to underpin that portfolio of initiatives? >> It's a a great question and some remarkable results, Preina. So where business cases and initiatives were very much business focused, tech sort of followed and then we went through a position where tech followed and then business was behind. We're now seeing where both are now in the room and I I mean we we hadn't prepped on this uh point whatsoever. But the point about having 10 people in the room, two engineers, eight from the business is actually a very meaningful ratio. So the two engineers are probably all AIDed up. Um so they are going into that session having claude or something with them as engineers. Um you got business people who understand the pain problem which is could be with a client could be regulatory it could be whatever the pain problem that he or she's trying to solve and they are now working together to go and deliver a solution or a proof of concept or solution enterprisewide for an insurer. What I find fascinating through this now is the word trust. Um, garbage in, garbage out, um, was a cliche that I learned very much in 24 years ago in studying AI back at university here in the UK. Um, garbage in, garbage out is as important it was 25, 30 years ago as it is today. the ability to bring together the both the internal and that external data together. The ability for the organization to have faith, accountability, transparency and trust on that contextualized data is so critical, so critical to get the value out of this once-in-a-lifetime uh technology set. And if I see organizations that we support the likes of city as well as many in insurance as well as government if you look at in within government per se lots of fragmented data but once you can bring that data on staying on the right side of history as you bring that data together then apply the best of machine learning and AI on top of that curated trusted view of data the results are fantastic and we see that time and time again. So it's very exciting but the key thing that I really want to underscore on is trust in data. So a portfolio of initiatives that are increasingly impressive being scaled underpinned by a layer of trust right. So David coming back to you. So as we see those in initiatives scaling and being embedded within critical workflows, how do the governance models and the risk models keep pace with that change particularly in a global financial institution such as city? >> Yes. Um I mean I think Vish said it trust is incredibly important because without trust you know worst case scenario you can cause a serious issue for your customers but actually even if internally if people don't trust the system they won't use it. You will not have impact you will not have adoption. So you do have to make sure that everything that you've done is um completely explainable. So we talk a lot in this era of AI about evaluations. So tests that a system must pass before you can trust it and give it to a user. Those are incredibly important. So one of the things that we do is center all of our discussion and explanation around our AI systems on the level of evaluation that we have in place. Now for us in city um we are a very globally scaled organization. We operate in over 160 countries in the ground in 90 countries. We have trading floors in 77 countries which is a stat I love. Um and at our scale complexity is the enemy. So we've been very very intentional about the technology design, how we implement technology controls, how we explain those controls, but just as importantly, we've been very intentional with how we educate our people, and how we inform um not only why we're doing the things that we're doing, but what they are, how to use them, how to discover them. So um very very machine-like in terms of the dis the creation of the technology and then very very methodical in terms of how we educate around the appropriate usage of that technology but at the heart of it is that technology we spent a lot of time on inbuilt guard rails that stop the technology going wrong um from the outset. So, we've um launched a platform called Arc recently, which is an agentic runtime. And what's most interesting about ARC is all of the things that you would want controlled to avoid these horror stories you read in the media about agents going haywire or they're all wired into our platform. So, we built that first. Um, you know, it would have been a disaster for us if we'd have rolled out agents everywhere and then thought, hang on, we probably need a control layer and tried to retrofit it. So, being very intentional, I think, is the key word for us. Um, and it works because we've scaled very rapidly. Um, we have over 190,000 users of our core AI tools at the moment. Over 80% adoption, very high engagement level. We have over 30,000 developers who use um AI coding tools and over 11,000 of them operate an entirely aentic software development stack. So, we've had some really meaningful adoption and that is turning into meaningful numbers. Like we're far beyond the experimentation phase. We've definitely seen hundreds of millions of dollars a year in benefit from our developer AI work alone and much more to come. >> Some impressive numbers in terms of scale and clearly trust and controls really a a foundational guardrail around this technology. But you mentioned the benefits. I'd like to move to the value case for AI. While trust is important without the value case and the business benefits the board is not going to invest further into the technology. So as you think about this kind of benefits case in the context of city that benefit typically doesn't appear as a um easy to account new revenue line. It's going to be a combination of productivity speed risk reduction and other qualitative benefits. So how do you build the measurement discipline to be able to track those benefits in a way that is um uh accepted by both the business but also the risk governance committee? >> Yeah, it's a great great question. I mean there are three classes of benefits very simple. It's revenue, it's expense or it's an improved control environment. Um so on the revenue and expense points we have um a very disciplined approach with our CFO. So before we engage on any initiative I mean I think as Pana says there has to be a very deep business case. It's actually quite painful to put together and it puts people off who won't commit. You know if they if we ask a team um or a team have asked for investment in AI build and they're not prepared to put an efficiency in their 12 month forward are they serious? So we challenge and challenge and challenge. >> Um and so because we've been very disciplined about that, we've got a very standardized approach about how we capture benefits, whether they are expense and which most of the benefits have been on top of the risk and control benefits. Um but now we're increasingly seeing the revenue cases come in. So um you know things like the developer benefits that we've had so far are ultimately less dollars for the same or much more output. It is an efficiency. Um all the work that we've done in our services business and our loans underwriting business where we're seeing four times the throughput, 50% less manual work, those all efficiency benefits. But now we are moving into the era of it driving revenue. So we did um we've had a wonderful partnership with Google Deep Mind on a product called Google Sky, which is a wealth avatar, a virtual one. Um it's very very impressive to look at but what's very interesting to me is all of the engineering behind it in terms of surfacing the right data so that um Sky can have a high quality dialogue with a client. It will drive more revenue. Um but what's very exciting for us is that um sort of AI to support a productive client relationship applies in many parts of our bank. It's highly cross-licable approach. Uh and then there are things like we're doing within our equity research team in our wholesale bank about how you um process economic world news change put into financial models create content distribute it highly optimizable through AI and so we're working on some things like this too which we're we're pretty excited about the um revenue ambitions on that side. So lots of benefits, lots of use cases and lots of excitement throughout this industry and many industries around the potential benefits of AI counterbalanced by also some fears and fears around the impact on people humans. Right? So Vish, I'd like to go to you next. So this as we think about the impact of of AI, do you think that the AI industry has an obligation to build AI systems that genuinely augment human expertise rather than just displace it? >> Yes. Simply um as technology leaders as we are um it's not a sidebar responsibility around people. It's at the core and at the center to how we build technology, take technology to market, service our clients. The human aspect is foundational to staying on the right side of history. As I mentioned earlier this morning, um there is a training piece here. There is an education and a training and an adoption piece. David underscored on the adoption piece. How do you take people on the journey? Now, here in the UK and in Britain more more generally, we can't push this technology away. We will lose competitive advantage if we push this technology away. So, we have to bring this in. And there is a piece around sovereignty around the data and at the AI in particular when it comes down to distress to the system around that sovereignty piece. So sovereignty under pressure is obviously a critical point but the human plays a critical role here a very critical role. Just looking at my organization. So we're about a thousand people now worldwide in 19 countries and we have rolled out AI right across the business who are far smaller than city. Our adoption rate is you know it's in the high 80%. In the organization if it's across finance people or HR in R&D is at 100%. So it's a mix between function but overall we're over and the efficiency and the effectiveness that we get out of this technology is foundational to the way we run and operate as a business but it's at the human coming first. So when we when we deploy our technology in our clients I'm very very forthcoming around having the human still in the loop on that journey. Now there may be certain cases where human doesn't necessarily need to be in the loop with the right framework and guardrails around the technology. If you trust the data, you understand the foundations of the model, you understand the transparency of the model, then in certain cases the human may or may not be in the loop. But that needs to be documented, needs to be transparent, and you're fundamentally taking a non-deterministic model and trying to make it quite deterministic. So that is a art and a science coming together very much. So the to underline your question, I could have answered it in 5 seconds by saying yes, but just to give it more color. The human is core to this uh journey. It is an enablement and a training and an adoption, but we mustn't as a society here in the UK push this away. We have to bring it in with the right guard rails and framework around the technology. >> So the human is caught to this, but we all accept that this is a huge wave and there's change coming. So David, from a a City Bank and a global banking perspective, what kind of jobs or activities do you see changing the most and how do you support that change from a people upscaling perspective? >> Yeah, I mean the genie is definitely out of the bottle, right? And I think to Vic's point, you know, it's out of the bottle for everyone in this room and it's certainly out of the bottle for our kids and everyone at university right now. AI is here and we've got to work out how to work with it and get the most out of it. So we really embrace that. So um one of our stated goals is to actually make our workforce feel like the most AI empowered workforce in the world with access to the best tools, best training, best capabilities. Um because you know we recognize that the future is an AI first future. So um so critically important um you know thing for us to focus on in terms of the kinds of work. So I think we see we don't think about jobs, we think about tasks. >> And so I would say there are definitely tasks that an AI can perform today that five years ago an AI couldn't and you would have had to have a human to do it. It's very logical to say that what if I had hired a person to spend 100% of their time doing tasks that an AI could do, I would find something else for that person to do. And that's really what is happening. So where you have tasks that are highly repeatable, automatable, there's a lot of coding work now. If you're in a call center doing lots of client dialogue, call transcripts processing where AI can do an increase in amount of that, those roles are going to change. They will not exist in the same form in the future that they do now. That's just the reality of it. But also I would say um and this is you know the really interesting point. um we are as a you know human race going to not sit around doing nothing right whenever we've introduced technology into the workforce like I can send an email now I don't need to write it down and put it in an envelope and put it in a post box it didn't mean we all fell out of employment it didn't mean all accountants fell out of employment when we invented Excel and you know with computer systems to do it um so I think a lot of the fear that we have at the moment is we don't know what the future will look like but there will be one um and the exciting thing for us all at the moment is we're defining that what it will look like. >> So our workforce will change in terms of what do they do but we're really looking to enhance and amplify the productive client relationships we have. If you go to any technology team they'll say great whereas last year I could do 10 tasks and now I can do a hundred. Well guess what there's a thousand in their backlog right and there always has been. There's always more to do than we've been able to do. We just hope this is a massive accelerant of like value creation for our company and all of our clients. >> So change coming but an optimistic lens about the future of work and what that looks like. >> Very much so. >> Provena, can I ask you to provide an insurance industry perspective on this topic? >> Yeah. No, absolutely. I mean I see everything that Dave is describing. Um but if I go back to the original question around how do you really drive value from this technology because people see the technology they see the process they see this is really two tangible things that we can change the thing that gets underestimated but if you get wrong is a thing that will make sure that you don't drive the value is the change management and the people engagement so if you think about implementing anything end to end probably 10% 20% is going to be the tech I I would say as much as 60% is going to be people related and that goes everything from not only just teaching somebody how to use a new tool or a new process but taking away the fear because everybody will read what comes around and will automatically think what's going to happen to my job. So there is it's it's on us to remove that fear away and the way we do it is we actually spend a lot of time on the people side of the change. So you know years ago we gave everybody the productivity tools. Now going back to the business benefit case sometimes and actually most of the times you do need a business benefits uh around any implementation. But when we rolled out the productivity tools we didn't do a business case. We made a strategic decision. You all know how much they cost. Just give it to the entire organization. 12,000 people were across 68 countries and let's see what happens. And it was fascinating. The curiosity, the energy, the oh, I didn't know it could do this created this ground swell of movement where people started creating their own agents and creating tools that help them on their day-to-day basis. So just making this technology available to people where it really matters at the front end, no matter what job they do, and don't spend six months figuring out a business case. It it worked. So that's the first thing I would say is you know be pragmatic about where you use it. The the second thing is around bringing people along in the journey and you will see a spike. You will see like this huge curve of adoption and then it will plateau because it'll be like well it didn't do what I wanted it to do and actually I've always done my job like this for the last 15 years so I'm going to go back. So keeping that momentum keeping that training and education is really key. Um so I think those are the couple of things I would say. Now, you know, you only have to look back at history and sitting in this room, we're surrounded by it. But you look at the industrial revolution, you looked at automation in manufacturing plants, the jobs didn't go away. The jobs became different. The tasks became different. And we see exactly the same thing. You know, nobody ever says, "I've done everything I came in to do today." And that's where we're seeing a real impact where we are moving away from making sense out of unstructured structured data. Um we invested over the last eight years of a strong data platform and data governance. And now we're moving away from that and making sense of it and making better decisions and augmenting the experts that we have. So that's the pivot we've really seen. But without taking our colleagues and our talents on the journey, that would be really hard. taking people on the journey. So as we approach the the Q&A, I want to ask a question to each of the panel. I want you to answer it in sequence. Um so starting with David. So thinking about our audience and the sea suite in the audience, what would your advice be in terms of the one thing to do differently over the next 12 months around this topic? I would say be incredibly intentional um about what you're going to do and um do it quickly. Um don't wait. Don't come up with a three-year transformation plan because by the time you've written it, it will be out of date. >> Um so be very clear, precise, keep it simple, do it quickly. >> Totally agreed about a three-year transformation. What David says. Um I just want to come to a separate point to answer your question is something what Preina said. We talked a lot about qualitative uh quantitative metrics, number of hours, revenue, etc. But coming back to the human side, you don't know what you don't know. And in many of these deployments, we've ended up finding stuff within data and AI that has absolutely changed the game when it comes down to managing risk and opportunities. So open-mindedness as you approach your programs, your transformation and going very much from business objective to data quickly, not in sort of um you know, you do your business requirements, then you do your project, then you do your testing. You've got to be more agile in the way you do this. And things that you think will take you six or nine months, it's taking you weeks now. So quick fast deployment more importantly not just focusing on the quantitative but there's going to be a bunch of qualitative that you need your teams to be listening as they deploy this those will end up being incremental business cases to your business case uh and that's one thing we're seeing more and more as we deploy um our technologies with AI or generative AI in firms there's so much qualitative that's coming through as well >> make this topic as part of your core business strategy, not something that sits in the basement or is the last thing on a long list of things that you've got to do. >> Great. Thank you everyone. So, we now move into Q&A. So, um you can scan and submit a question via Slido. So, we have a couple of questions here to get us uh started. So I I'll start with you provin. So question is why are fragmented pilots proving so persistent and what distinguishes organizations that are successfully scaling AI across the business. >> I think like with many organizations fragmentation everybody goes I've had a great idea and then you get busy doing what you're doing. So there's a natural inherent thing with humans of I need to get this done so I'm not going to share. And it's not from poor intention. It's just driven by a lack of communication. So I think it's um stepping out your usual silo boundaries of the way of working and instead of thinking in verticals think in horizontals of how your business operates. Um I think the other thing is it is difficult. Um, but embrace the ambiguity, embrace the this is going to be difficult and start because if you don't start, you're never going to realize the value. So I think it's thinking about business constructs, thinking about the way we run our businesses differently than we've done in the past. >> Any other comments from the panel? Yeah, I mean I think for fragmentation if you take hundred different approaches versus one where you have a hundred different risk and control conversations to have you have a hundred different potentially vendor conversations to manage. So again you acrewue complexity which is the enemy of progress and scale. >> Um so yeah >> Mish anything to add >> I was going to just echo the cross divisional alignment. um it needs to happen if it's from the claims office to the underwriting office to the data office that cross functional alignment which covers your people process and technology becomes really critical. >> Okay. Um right more questions from uh the audience. So Bish could you talk about minimizing bias and hallucinations in AI and what efforts do you take around that topic? Great, great question. Um, so the ability to ground an SLM or an LLM in data is a non-trivial task. I think you could probably echo that. >> It's hard. >> It's hard. Um, and first and foremost, the ability to take a model behind the firewall is equally as hard. Um now the way to govern this and what we have seen is pure sort of model risk management of the old has to be tweaked and turned to embrace the new world of AI. So in many areas where we have deployed within financial crime in particular things like AML transaction monitoring and so on the ability to do automated below the line testing the ability to understand sample groups as you're looking at risk and then matching it to a known data set is still a very important process. But more importantly in the world of AI and large language models it is critical. If you ask the question ever so slightly, ever so kinder, you will get two different outputs. So how do you govern that? So what we have seen and many of the LLMs and the deployments to date has been very much involved in the actual investigation not necessarily in the model. So the model of determining is Pina sorry Pina I'm going to use you is high risk and David is low risk. That type of conversation is yet to come and it's coming. But that is taking the regulator, taking the process and the policy, the risk frameworks that organizations live in, bringing that together. It will come. Right now, a lot of that generative AI deployment is more in the investigations and it will end up coming into the detection, but it needs to be treated carefully. Really need to be treated carefully. >> Yeah. I mean, I think um and I mentioned this need for evaluations and that's exactly what what they are. So it's this whole set of tests that you always run before you deploy AI into any real world scenario. So for example, we have AI that supports our US consumer cards business where we have over 100 million customers. We have AI in those call centers and the breadth of the tests that we run through um to make sure we don't have any conversations around buyers or protected characteristics to make sure you're always having high quality output. Um it's an incredible amount of testing that we you necessarily need to do and is appropriate. um before you do put anything um in front of anything like a customer or internally like good vibes is not sufficient, right? Um you you really have to have incredible rigor around all of this >> and again comes back to the trusted data asset. If you are inherently taking a biased data set into a model, you're going to get a biased set of results. So it's got to be trusted, it's got to be governed, it's got to be curated, it's got to be contextualized with the right lineage around it. This becomes really critical as we go into the world where AI meeting data for outcome becomes reality. >> Yeah. And I think you'd also have to make this operational. So it can't be something that sits in a a strategy or a standard. So the way we think about it is control and governance by design and in by default and in depth. So you always so often you often think about these things when you're building them and you then you launch them but it has to continue throughout the whole life cycle. So it's the regular checking, the regular monitoring, making sure your data assets are owned and governed and your data quality indicators are managed and monitored. Um, so I think goes with the entire life cycle because ultimately it's about digital trust. This will become another topic for digital trust and what we're also seeing which I think is brilliant. It's been happening for a while is the role of technologists and the business are becoming closer and closer which then helps the spirit of governance. Um and I call it being bilingual because there will come a time where you won't know who is a technologist and who is a business person and I think this transformation we are going through will only accelerate that journey. >> I think we have time for one more question. Um so let's see. So, Praina um how do you or how did you decide on how much AI governance is enough so it doesn't slow you down but at the same time it's safe particularly as regulators are still finding their own way >> absolutely it's a question I get all the time and you can imagine from the practitioners they're like we just need to get this built but then we're going no but we have to do it in a control way and you have to be pragmatic IC but realistic as well and make risk based decisions. So we did operational things like we took our you know every IT division has a control process. We looked our existing control processes and just updated them to include this new piece of technology that behaves in a different way to other things that we've done in the past. And then we've take we when you look at AI you look at two factors. How much agency is it going to have and how much autonomy is it going to have? And based on that you say okay what level of controls do you need to put in place. So it's thinking about what is it that you're going to implement um when are you going to implement it and what can the impact and the risk of that be and then totally reverse engineering the governance around it. So where we are today we are augmenting our decision makers and augmenting underwriting. We are not replacing human in the loop is super critical. Um, so those are, you know, the how we work, but then it's enhancing your existing processes and monitoring how long controls and things take because if the control process takes way longer than the thing you built, it's an indication of you're out of kilter. So measuring and continually improving both the development life cycle, the launch cycle, and the control gates is really important because only then will you know if it's out of whack. >> David, anything to add? Yeah, I think um one thing that we found is as quickly as possible being very specific about AI risk. Um to start with what's the right level of governance or control, you've got to start with what is the risk. So um in our we've got a very sort of robust risk risk taxonomy across all of our businesses, both operational data, you know, several other factors. So we're now layering AI risk into that taxonomy. Um when you have that view, you can then start to think about process risk where you're using AI. You then think about the necessary level of control. Um and everything flows from there. If you can't quantify um you know where risk is where your AI risk is too much or too little then how do you know? So you have to start with being very precise on an understanding and definition of AI risk >> and there a different perspective from you as you're on the service provider. >> Yeah. I mean it's it is a balance between the risk and the opportunity just like you know a manual car you don't want to spin the car you don't want to stall the car uh you want to continue driving forward um we all understand the opportunity here but coming back to my first point staying on the right side of history on this technology is going to be really critical for society um but looking at the advantages you know one of our clients who deployed our platform was looking uh to detect uh financial crime crime in in a particular area around AML. We uncovered a large gang of human traffickers using the best of machine learning and deep learning and AI here. Now that to society is critical. Detecting disrupting a human trafficking sex gang here in the United Kingdom was a big qualitative it started with but then when you look at this network of individuals and shell companies was a huge problem for society. So opportunity, risk, got to take the balance. But let's not look at the big goal here. The big goal here, it's a transformative technology that used correctly will do tremendous for our society. >> Well, I think we're getting to to a close. Um, as I reflect on the discussion today, we talked about a transformational once in a generation technology that will have a a meaningful impact on our industry and many of the industries with a lot of risks to be managed but at the same time a lot of opportunity both to have business benefits but also to uplift humans in terms of the work that they do, the nature of the work and also uplift societ. society in general in terms of being able to um disrupt uh criminals and and sex gangs, etc., which I'm sure we would all support. So, please join me in thanking our panelists.