Andrew Holdway of Swiss Re on the operational impact of fixing insurance intake

Speaker A: Welcome to Unstructured Unlocked, a podcast where listeners discover how insurers are entering the decision era, utilizing artificial intelligence to refine their decision making processes, boost underwriting profitability and achieve premium growth. Welcome to another episode of Unstructured Unlocked. I'm your co host, Tom Wilde.

Speaker B: I'm your co host Parol Colegreen Barw.

Speaker A: We're really excited today, uh, to welcome a great guest for a very important topic around data augmentation, um, as it relates to risk selection and underwriting. Andrew, uh, uh, Holdway joins us, head of partnerships from Swiss Re Risk Data Solutions. So Andrew, welcome to the show.

Speaker C: Yeah, hi, thanks Tom. Thanks Parul. Thanks for having me on the show. Um, a quick introduction of where I'm coming from within the big Swiss Re group. So obviously Swiss Re, well known reinsurance and insurance brand, 160 years old. Um, and where I sit is a business unit called Risk Data Solutions. And what we do is effectively take that 160 years of risk knowledge data, uh, insights, repackage it and then present it back to the market for our clients and prospects and other business partners to also consume and use the same insights that we use to run our uh, reinsurance, uh, and insurance business.

Speaker B: Brilliant. So let me kick off with uh, the first question, Andrew. So insurers have access to more first party, second party and third party data than ever. Why is it so difficult to get visibility into this data, uh, early in the process?

Speaker C: For me there's an underlying theme of legacy, legacy processes, legacy systems, legacy behaviors. We still see also some fragmentation between business units in the same organization where insights and data points could be recycled into different workflows in the value chain. But quite often they remain stagnant in a data repository for single use. Um, unpacking the legacy processing systems. Well we could spend a few hours talking about this, but what we hear from insurers is that whilst they may have access to data somewhere in their business technology capabilities, to be able to consume APIs and therefore, uh, consume and use decision intelligence consistently at relevant points of decision is lacking. And even when there are capabilities, trying to get integration prioritised amongst an already packed tech and data roadmap can also be a challenge. And when I mention behaviors, it could be that silo setup within the organization. But more worrying for me is that people's behaviors are different which leads to inconsistency of decision and action. Some data used here, some there, some with manual input and potential for human error, some automated and therefore as an output in terms of underwriting decisions and Then eventually portfolio management, it's difficult to identify and then measure what data was used to derive both positive and negative results.

Speaker A: That's a great description. It's kind of the first order problem, which is how do you get that data as part of the ingestion in the first place? Is there a second order problem? I mean what happens after that ingestion moment? Does the data sort of break apart from the ingestion? Is that a challenge that insurers face in leveraging the data further into the process?

Speaker C: I think about process automation, enrichment and again consistency of use. And I think we should also think about the step before ingestion as this is where the data journey begins. What format is the data provided in? What's the quality and quantity of the data? Is there a mix of structured and unstructured data? How reliable is it? What's the use case workflows it can power from a process point of view. Quite often we see data being ingested for one specific use case or one specific part of the workflow and not passed down or recycled further down the value chain. Take triaging for example. Sovs are sent in by a broker and there's a data enrichment to prioritize risk based on appetite. But once that happens, is the data then also reused for risk analysis, pricing, accumulation, management, portfolio steering risk? Rarely. And if it is, is it being used consistently? Is there a manual process or is it automated throughout the decision making process that results in a risk acceptance decision that also creates immediate visibility to other stakeholders to also take advantage of that decision intelligence which was received and enriched at the start of the process. As an example, I spoke with a global insurance carrier last year and to a team responsible for portfolio management. And their pain point was that uh, they had to make decisions based on performance metrics when they knew there was inconsistent use of data and in many cases several sources of inconsistent data across the value chain. They were fine with secondary sources of data to give a second opinion, but in an ideal world they were looking for consistent data being used from ingestion through underwriting and policy bind to be more comfortable in making portfolios team decisions

Speaker B: with confidence on contextualization of data. Andrew, how does the lack of or having context around data impact underwriting and risk analysis workflow on a day to day basis?

Speaker C: Okay, so not a lot surprises me these days. I mean from geopolitics to natural catastrophes to even my own football team results. However, I would be highly surprised if those responsible for underwriting and risk analysis do not have full context. The Data they are using. If they are using data blindly or even remotely blindly to make decisions, they should be hung, drawn and quartered. But what I think is important here about context is understanding the depth, the reliability, the limitations of the data so that informed decisions can be made based on the data at hand. Uh, how credible is the data? What's the source? What's the science and methodology behind the data? How often is data updated? Where is the data strong and um, not so strong? Could it be at peril and hazard level, geographic level, capability and use case level? Knowing the blind spots is important, just as knowing whether data differentiates and also the confidence level of the data too. We talked earlier about consistency of use of the data, uh, impacting underwriting workflows and risk analysis and how the more automated, the higher degree of consistency it will be used for. Now, I think we're all aware that for the more commodity and volume risks in business, the aim would be to automate triage and underwriting based on business rules and appetite. And therefore, in theory, data being used consistently within frameworks across the value chain results in performance being easily measured and the business rules being set with input around the data being used. If the context of the data here is misunderstood, a carrier may be writing a large volume of business, which could result in a nasty surprise and to later prune a portfolio not only as an operational expense, but a reputational risk too. And on the flip side, the carrier could be declining business or being uncompetitive, which is actually in fact more attractive.

Speaker A: The CAD data is typically available very early. Right. As soon as you have a location, you are able to start to assemble, you know, a picture of what that risk looks like? Um, do you find that carriers, you know, are using it high enough in the ingestion work stream? And what advantages emerge, um, when they, when they do, you know, when they look at it kind of at that moment of ingestion.

Speaker C: Yeah, there's, there's no one size fits all here. I think all carriers are using data at different parts of the journey. Some are more mature than others, but clearly that there are advantages of bringing that data risk data upstream. I mean, the clear advantages here is efficiency in terms of speed in decision making and servicing brokers and business partners, and then underwriting and operational expense. I came from a carrier before I joined Swiss Fri, and carriers do spend a lot of time trying to educate their business partners about what their preferred risk appetite is. But we do not live in a perfect world where an sob with all 10,000 locations hits the bullseye plus business partners change, people forget. And carriers risk appetite also can be fluid too. When I first started out in insurance and we received a mix of faxes, paper submissions and emails, it was just impossible for underwriters to prioritize what business they would be looking at. First it was literally reviewing each submission slip and sov as and when it landed on their desk. And I even used to try and influence their prioritisation with donuts and cookies so they would work on my broker business first. Uh, now having the data enriching submission slips at the time of receipt, automated triage based on risk appetite and other factors leading to underwriters open up their inbox in the morning. And having a defined priority list creates focus and discipline. It creates focus on where they should be spending their time but equally as important creates visibility on where not to as well. And these info points can be used to communicate with business partners internally and externally and manage expectations and the output. That triaging stage can also be used to reflect on risk appetite and source of submission to then also help re educate and or reassess portfolio steering direction and having the data really to back it up.

Speaker B: So Andrew, tell me what really changes if you're doing risk analysis and data enrichment earlier in the process rather than much later? What is the ROI or operational efficiency you gain by doing so?

Speaker C: Yeah, well as I mentioned previously, for me it really creates that operational domino effect where more focus, more discipline, more effective and efficient decisions can be made. It really is win win for the carrier and the business partner. It's really highly frustrating when efforts are exerted at the beginning of a risk analysis process only to get to the final stages and the underwriting process is delayed or even eventually stopped. These kind of issues should be nipped in the bud as early as possible in the workflow. And that's why bringing all that data upstream really then has that domino effect further downstream. When half of those issues have been put to bed. There's prioritisation to teams to focus on and ultimately the teams are spending more time on the business that they want to write. You've also got the doing more with the same resources or doing more with and less resources. But really it's that quicker, better decisions based on that triaging at the top and downstream outputs where everybody knows where they should be working with the credible data to make those decisions.

Speaker A: Yeah, let's dig into that a little bit more. So I think that you made a couple key points here. Use the data early in the process. Number one and two, keep that data bound to the submission throughout its life cycle. Um, if you execute on those two things, what are some of the sort of second and third order benefits that ah, that you see and where do they land across the various um, teams that participate in that side of the business, underwriting actuarial ops, et cetera?

Speaker C: I think when you look at underwriting actuarial operations that credible consistent data is used upstream, then in their individual universes, in their remits, they all benefit. But if they're all in sync with the consistency and quality of data, then they all win together. Sharper pricing, better underwriting results, more efficient processes protecting the profitable base portfolio and driving attractive new business growth. Everyone wins. In these circumstances, if each group has the understanding of the other in terms of when, how and why the data is being used, then this naturally drives confidence and trust between teams. I've worked in organisations before where there can be finger pointing if there is collective underperformance which then erodes trust. Underwriters not closing business because the prices are too high, brokers not being serviced because turnaround times are too long and actuaries unable to price competitively due to lack of data. If all are on the same page again and using consistent and quality data from step one, this alleviates many of those pain points and then having the data to reflect and analyze each team and um, the collective performance allows for better steering of the business and improvements to be made along the way.

Speaker B: So the market is soft, not much rate is getting pushed into the Premier, which means that insurance leaders are increasingly focused on how to make their processes more efficient and improve operational expenditure which is stubbornly high. What's the most practical way to improve this upstream pre bind data support in core operational workflows without actually pushing up the opex, uh, of an insurer, a carrier or a cedent in your case?

Speaker C: Yeah bro, if I had that magic bullet I would certainly be a rich person now. I mean, I think the first step really, and we have these conversations with clients very regularly, is do they have a clear view of the data landscape Today, um, we see business units operating in silos, different lines of business. Is it retail, then is it commercial, um, then the different teams of underwriting actuarial operations. So I think that first step um, for businesses is really to get a clear view of that data landscape. What data is coming in, where, when and for what use cases and decisions really starting and having this data map should give visibility regarding any duplication of data quality of the data for the specific use cases, but also highlight which workflow decisions are being made where there could be either insufficient data or where data could be used more effectively earlier in the workflow. Once this is clear, um, testing and pilots could be an immediate step to start reimagining data inputs and really then to start measuring output versus the BAU environment. Today what we see more and more as you said before, clients really wanting to test the business decision output, uh having a clear idea of the roi, making any change in terms of process and or data input and really any change or influence on core operational workflows sounds like uh, an enterprise level shift which would need various business cases approvals. Therefore starting smaller and improving the value of having that upstream data before going big bang is a really common practical uh way to do this. It's cheaper, it's quicker, it's not going to have to go at the enterprise level. It could start smaller at certain business units. Um, but like with any test or pilot, really critical to be very clear on what can be measured and what success could look like and within a clear time frame this would be a good starting point on the journey of using quality and relevant data upstream for downstream benefits and really being able to measure that tangible benefit.

Speaker A: Well Parul, we can't let Andrew escape without talking uh, about AI and the impact on uh, Swiss uh, RE's uh risk data solutions, product development, roadmap the future. What do you see Andrew in your world in terms of where AI has had the biggest benefit and maybe where are some of the risks in applying AI? I mean you live in the predictive analytics world very squarely Tom.

Speaker C: I'm surprised it took so long to mention AI in the conversation but yeah, look the SwissWeek group has um, like most large organizations now are really reviewing where AI can bring value internally but also when we go externally as well. And we've done a lot of work around our own operational efficiency and how we can use AI internally. I think we're moving cautiously towards how that then can be used on our external propositions. There's a lot of governance around AI and where we can and where we can't, where we should and where we shouldn't be using AI for specific use cases. Using AI for predictive analysis analytics is always um, a sensitive topic in a fast moving natcat environment that we are in today. We do partner with uh, Google X one of their business units called Bellwether, where we look at our wildfire risk, where we look at a one year and a five year view of how that wildfire risk could spread. So Predicted is always a tricky one. But certainly I think our initial use cases are more uh, in our internal operations. But we have a very, very clear direction in saying, well it's not just internally that's great things, but how can AI really value, bring value to our customers and our clients? And I think that's on the horizon next in the Swiss RE roadmap.

Speaker B: That's brilliant. I have a quick question now I've worked in insurance industry like you for 20 years. Do you think that incrementalism in any industry, AI or technology adoption is a curse, uh, of insurance industry or really a boon? Because we are thoughtful, do you think we miss trends or are slower to get behind trends and that means we realize growth potential of a technology later rather than sooner?

Speaker C: I'm going to answer that in two ways. I think there is a desire to adopt certain technology and solutions quicker than businesses can. I mean there's already technology roadmaps, there's always the anchor of legacy operations that restricts speed and pace of adoption of these new capabilities. But I also think, um, there's an issue around trying to adopt AI for AI sake. It's a buzzword, it's everywhere. And we've spoken to business partners and clients and before asking about what solutions we have and the benefits to their business. The first question is do you have anything AI? Whereas for us AI is an enabler. It's not the solution, it's not the risk insights, it's an enabler to deliver, enabler to assess. Um, so as a broad answer, I think we have anchors in being able to adopt solutions, not just AI but for other solutions. There's still a big movement from on premises solutions to the cloud. I think obviously when we get to the cloud, those AI solutions become a bit more agile and more easy to deploy. But I would also give words of encouragement to colleagues out there that making sure that we're using AI for the right reasons, for the right business case, for the right use case, and making sure that we're deploying it in a governed, reasonable manner without trying to just go big bang, it's AI, let's adopt it because it's AI and then not have the risk controls in place to manage it and to control it. So really I think um, it's an adoption point of view operationally, but really finding the right use cases. And I honestly don't think we're all there yet in understanding the true value of AI. I think we're all trying to run before we can walk because it's the buzzword of we should all be adopting it, but do we know how, do we know what for and do we know when we should be doing it?

Speaker A: Maybe as a tangent to build on that. Um, but also, um, switching gears slightly, I think in the news there's an awful lot about insurers pulling out of regions of the world um, because of catastrophe risk. If you think about certainly wildfires in various parts of the world, California, Australia, et cetera. But counterintuitively, does better predictive ability and better data actually unlock insurers ability to write risks if they can be more precise about where the risk really lies rather than saying we're not going to write in Florida, which is a big blunt instrument. Have you seen cases where no, that's not the right response? The response is let's be more precise about where the risk lies so we can properly price.

Speaker C: Yeah, absolutely. And the California wildfires last year is a great example where we, we looked in quite granular detail on the maps of locations within 10 to 20 yards of each other. Where in a large pixelated area you might just think that's not a good risk. On the other side you might think it is a good risk. But when you look more detailed into not just the lat long location of a property, but the building characteristics, the exposure of that property as well, the elevation of it. When we look at risk, uh, mitigation measures, is it protected or is it not protected? Does it have bush or uh, vegetation close or not? So having more granular risk insights, not just on the NatCap peril itself and exposure to that location, but having a more insight into the property and the exposure and the vulnerability of that property as well should give underwriters a better view of the overall risk and not just the overall risk today. But what could be that risk? With the climate change and the development, what could the risk be in five years? What we see helping carriers, having more granular risk insight and the vulnerability data is looking at underwriting today but pricing for the future. Something like Wildfire is a very fast evolving um, risk that we have. And to continue, for example, just increasing rates 50% every year, it's not going to help everyone. So how do we price for the future? By underwriting today and making sure there's a softer land on that five year premium if somebody buys a policy today. So I think there's a number of benefits on the underwriting side, on the pricing side, on the accumulation management side as well is critical. We worked with some carriers last year where we helped to manage their accumulation or how they viewed accumulation when they were underwriting risk. And that allowed them to decline certain risk or accept certain risks on certain conditions. So great benefits to the carriers. But I think there's also a more uh, profound advantage to the actual policyholders as well where we can communicate to these policyholders to say actually you are in a high risk zone and therefore if you do adopt some of these mitigation measures, then not only will you protect your property in a better way, but then that could also have an impact on your insurance premiums, et cetera. So um, I think um, when we talk about tech and data and the benefits to the operations and the underwriting, I really feel we should also be communicating the value of that to the insured and the policyholders too. And I think that's currently lacking as well. I think something we need to think about more in more detail.

Speaker B: So do you know any insurers who are doing a good job of it? Because last year we had devastating fires in California and a lot of people were uh, left in alert. Their entire properties were destroyed and uh, they didn't have timely information. And most of the brunt of uh, um, actual uh, mitigation services and informing was with the local governments um, and disaster agencies. Where could uh, insurers come in and provide proactive information? Because we've got, we say to everyone that we have a social function, we are safety net. But we don't see them proactively providing this information which we do have. Do you know anybody who's actually doing a good job of it?

Speaker C: Yeah, I do actually. Swiss RE is doing a pretty good job if I can promote ourselves. We do have um, a team called Public Sector Solutions. Pss, uh, and what we do there, we help generate and benefit the public private partnerships. But we do work with government and public entities to provide some risk data, some risk management services. There's Canadian provinces that buy the data to look at Flood, for example. We have some examples over in APAC in Anz where uh, government entities are using our data to help them manage their risk portfolios. We work with Flood RE in the UK so and I'm sure we're not alone. I'm sure there's many of us out there working with these public entities. Again I'm imagining it's just a simple case of their own prioritization. The speed that uh, a public entity could move at the cost that we have of the data versus what they could get maybe publicly available. I won't comment on the political environment in The US and the cost measures that have been running through some of the government departments where all of a sudden there may be less funding to look at purchasing data or working with private entities. So I think, um, it's not an end omission, it's a journey that we all have to continuously working across public and private sectors to make sure that we're really again the value to the person who owns the property, how do we solve their risk management topics and then working backwards up the value chain to see where we can all play a part to do that.

Speaker A: This is great, maybe to bring us home here. Everyone likes sort of top three lists. So what are your top three recommendations for uh, carriers in better leveraging data, um, in their ecosystem and to capture the most benefits there. How would you, uh, prescribe?

Speaker C: Yeah, so I think if there's one person who owns the data repository, I'm sure there's many in organizations. First of all, it's really just to uh, try and step back out of the business and look at that data landscape understanding. Again, as I said before, what data is coming in from where, when and how is it being used and then really having a reflection on is that data still fit the purpose for today? Many data sources would have been around for several years, haven't been updated as the business has also changed their risk underwriting appetite or entered new lines of business or new markets. So as a first step, I'd encourage the teams to look at the data that they're using today and really being critical on is it still fit for purpose for the business that we have today compared to the license that we bought 10 years ago?

Speaker A: Perfect. Well, we've been talking to Andrew Holdway, the head of partnerships at Swiss Reed's Risk Data Solutions. I'm your co host Tom Wild.

Speaker B: I'm your co host Parul Cole Green.

Speaker A: And this has been another episode of Unstructured Unlocked. Thanks so much.

Speaker B: Thank you for listening. Thank you, Andrew.

Speaker C: Thank you.

Speaker B: Thank you for joining us for this episode of, um, Unstructured Unlocked. You can find all of our episodes wherever you listen to podcasts today, Spotify, Apple podcasts, anywhere.

Speaker C: Be sure to write a review if you like what you hear.