| If you don't have an account, Register here |
March 2, 2026
In this HFS interview, Genpact’s Yasir Andrabi and Macdonald Okolie discuss why AI has so far fallen short of expectations in insurance—and what needs to change. The conversation explores how agentic AI, data-driven underwriting, dynamic risk prioritization, and the Genpact Insurance Policy Suite are helping insurers move beyond basic workflow automation to fundamentally rethink underwriting judgment, capacity management, and underwriting economics.
You can listen above or watch the videocast here:
In the discussion, Yasir Andrabi, Agentic AI Leader for Insurance, and Macdonald Okolie, Global Head of Insurance Underwriting Practice at Genpact, outline a clear reason insurance AI hasn’t moved the needle: most insurers have digitized underwriting workflows without addressing deeper issues around data quality and decision-making.
Reviewing the Genpact Insurance Policy Suite, the conversation focuses on shifting underwriting from speed to judgment—using agentic, data-driven prioritization to direct limited underwriting capacity toward the risks that matter most, based on profitability, risk-adjusted returns, and likelihood of binding.
By embedding third-party and exposure data while maintaining clear human-in-the-loop controls for accountability and regulatory compliance, Yasir and Don frame underwriting AI not as another layer of automation, but as a structural lever for changing the economics of underwriting.
You can also listen on:
This transcript was auto-generated from the original recording and lightly edited for readability. We've done our best to catch errors, but names, technical terms, and company references may be misspelled or imperfectly captured. For the definitive version, please refer to the original audio or video. Views expressed are the speakers' own.
Welcome to the latest edition of HFS Unfiltered. I’m Tony Filippone, Head of Enterprise Advisory and Insurance Research at HFS Research. For more than two decades, the insurance industry has invested heavily in digitizing underwriting. Core platforms, workflow engines, analytic tools, yet underwriting outcomes have barely moved. Our research here at HFS shows underwriters still spend half their time on non-core administrative work, and about one in five insurers are confident in their ability to meaningfully innovate their underwriting. What’s striking is that this isn’t a lack of belief in AI or data. Nearly 90% of insurers expect real-time data-driven underwriting to replace traditional models within the next five years. Yet fewer than a third are structurally ready to support this. So this isn’t a technology gap. That’s an operating gap. And that’s why today I’m excited to be joined by Yasir and Don. We’re going to be talking about how AI can materially improve underwriting judgment. How risk is selected. How pricing decisions are made. How underwriters focus their time and work that actually drives profitability. So let’s get started first by asking Yasir and Don to introduce themselves. Yasir?
Thank you, Tony. It’s great to be here and thank you for having us. My name is Yasir Andrabi. I lead Agentic AI Solutions and Strategy for Genpact’s insurance business.
Good to see you again, Tony. The name is Macdonald Okolie. I lead the insurance underwriting practice for Genpact globally.
Super. So let’s move into the first question of the day. The insurance industry has spent, as I said, two decades digitizing underwriting, delving deep into core systems, workflow tools, analytics platforms. Companies have spent millions of dollars on core platforms, yet underwriters still spend nearly half their time on non-core work. At the same time, as I said, our research shows that only one in five insurers are confident in their ability to innovate underwriting in a meaningful way. From your vantage point, Yasir, what fundamentally went wrong with the industry’s digitization strategy?
See, I think, as you rightly said, insurers have spent decades modernizing where work happens, which is policy administration, workflow engines, rules, and analytics, but not how underwriting decisions are actually made. And the strategy assumed that if you made a process faster and data more available, better underwriting would naturally flow. And that didn’t happen because much of fundamental underwriting is not a linear workflow problem. It’s a judgment, prioritization and most importantly, a trade-off question. So just by providing more data or by providing cleaner, faster digitized processes didn’t help. And I think that is where it was challenging. One of my favorite analogies is picking up paper, taking a picture and saying, I’ve digitized it. You digitized it, but did it actually modernize? And what was the outcome of that?
And what problem do you think insurers think they were solving, but they weren’t addressing?
That’s a good one. I think most investments were solving for speed and scale, leaving out the decision clarity. And that’s where the gap happened in underwriting modernization. Now, the two bottlenecks that haven’t adequately been addressed over the years has been prioritization. So what deserves attention versus not? When human judgment adds value versus when it doesn’t. And one of the big things that we saw when RPA was at its peak was people were automating for efficiency, but unfortunately, automating a bad process just does a bad process faster. And it didn’t yield results. And that’s what happened. And lastly, how signals become decisions and not just static dashboards was another miss that we believe that caused underwriting subpar optimization in underwriting.
There’s such relationship-driven underwriting too at times where facts and data weren’t being used and actually underwriting, it was amazing to sit behind an underwriter and watch the art get done without a lot of science at times. And picking through the email, overflowing with submissions from brokers and figuring out what to underwrite or not, they almost lost the science in a focus on just trying to get done what they knew they could get done. It was shocking to me. Absolutely. You know, when I think about it, you know, most insurers still describe their underwriting as a process problem. Yet, you know, the data that we collect here at HFS shows that nearly 90% believe data-driven real-time underwriting will replace traditional models within just five years. Some of them are trying to do it faster. Fewer than a third are actually ready to support that shift according to our data as well. So, Yasir, you’ve been very clear in our conversations in the past and your messages to the market that underwriting is fundamentally a data problem. Why does that distinction matter, data versus process? And why does that change what AI needs to do in underwriting?
Tony, calling underwriting a process problem assumes that the answer is just better workflows. And that’s what I was talking about earlier as well, right? Faster handoffs, more automation, a better workflow. But it’s existing steps that you’re taking. And if that were true, the two decades of workflow engine optimizations would have solved that problem. And we know it hasn’t. It has optimized efficiencies, but it hasn’t solved the underwriting problem as such. So why underwriting we believe is actually a data problem because the hardest part is not moving work. It is making sense of the fragmented, inconsistent and incomplete information as real-time as possible so underwriters don’t struggle because the problem is unclear. The problem is very clear. They struggle because the signal to noise ratio is poor and that is what we have to improve fundamentally in that entire underwriting process, if you will, through data. And the distinction matters, you ask, why does it even matter, right? And I think the distinction matters because processes can be standardized, but you cannot assume that data is ready to be consumed. Your workflow efficiency doesn’t fix the latency, quality, or context gaps. You need data for that. And that is what drives better decisions, not just faster decisions. And I think that’s why it is very important to make that distinction, approach it with a lens of it’s a data problem to solve so that we can get better underwriting decisions. There was a third part in your question about why this changes and what AI must do in underwriting. Approaching the problem resolution with the data lens, we believe that AI must do something far more fundamental, which is curate and prioritize data before the decision is made. Resolve conflicts across the sources. And lastly, decide when human judgment is required and when it’s not. And I think that last part of it personally drives me very passionately because I think what separates tomorrow is going to be primarily how well those decisions of where to involve a human, for what to involve a human, versus where to let AI just go. And so that’s why I believe AI will actually be driving it in a better manner. So that difference is why belief in the future is high. But readiness is low, and that’s where insurers need help.
Yeah, it’s remarkable when you talk to underwriters out there, Yasir, and this is leading into a question I got for you, Don, they talk about efficiencies and they always think, well, it doesn’t make a difference if we have fewer or more underwriters. That’s a minor cost of the overall piece of the puzzle. If we write one bad risk, we can lose millions of dollars along the way. So it’s really about risk selection and pricing and adequacy along the way. It just changed the entire game from looking at process to how can you win your client’s work, you know, and the broker’s business along the way. Underwriters just simply don’t get paid to be fast, but they get paid to be right. Where does, you know, Don, where does the Insurance Policy Suite that Genpact has matured improve underwriting judgment, not just throughput or efficiency?
Tony, thanks for the question. Brilliant question. And I think it’s a perfect segue to the responses Yasir is giving. That lens of data first, that our insurance underwriting solution, Insurance Policy Suite, actually anchors data as its denominator. It does have four modules, and I’ll just touch on this differentiation and how it enables an underwriter to just be the super underwriter. The first module is around submission clearance. Now, put this in the context of seasonality, peak processing, and we know that most submissions that come to underwriters are abandoned. You make the point, their selection is based on relationship, not on science. The Policy Suite actually makes a selection based on science. Nothing is out. It has the bandwidth, it has the capacity, it will ingest every single submission, regardless of count, and it will in literally minutes give a point of view to an underwriter about whether this qualifies for him to take forward. The underwriter decides whether he wants to take it forward, but we know that brokers love underwriters that come back to them immediately, even if the answer is no. Right, so the underwriter is suddenly empowered to go back to the broker instantly with feedback on that submission and the broker knows where he stands because the broker is racing against time. They have to find coverage within a short period of time and they know they are now dealing with a trusted underwriter and that underwriter is not limited by volume. And then we move on to the next module which is the risk assessment module and this is all about understanding that risk so the Insurance Policy Suite will interact with the portfolio, it will deploy statistical modeling, it is taking this one submission, it is breaking it down into more than 500 rating factors and it is combing through the entire portfolio just looking for similar risks, it is trying to tell the underwriter something, here is where you have done this before, here is how risks of this nature have behaved, this will inform your decision, and it will perform several other checks. It will test for fraud, it will rate and score and I think one of the components that we think that certain lines of businesses will appreciate significantly is that it will also attempt to do loss modeling and extrapolation. So it will leverage trends, trends in the portfolio. In other words it takes the same point I made around let me understand what’s happening here and let me give you a point of view on what’s likely going to happen. So the point around bringing in agentic is that the data is there but the human eye cannot spot trends and so we use agentic to surface what a human might have missed and to surface this in real time. And then we move on to exposure management and all of this, by the way, Tony, we are attempting to deliver in real time if data is available. Now to Exposure Management, that module is not just geocoding, but it is also plugged into the portfolio. And at this point, it is looking at the portfolio capacity. It is trying to let the underwriter know the impact to your portfolio of this risk, the impact, any steps you need to take from a reinsurance standpoint. It will try to tell the underwriter if there is a concentration problem and it all boils down to brilliant underwriting, it is looking for that sweet spot that means this risk can be underwritten with minimum risk of adversity. And then lastly, it goes to the quote and bind module and we know this is the point of negotiation but an underwriter is negotiating just based on what he or she knows. But again, the Policy Suite is plugged into the portfolio, so it is making recommendations around here are things you might want to offer, here are deductibles you might want to suggest, here are clauses and inclusions you might want to take into account or remove because of what is happening in your portfolio right now. And so there’s that push and there is relativity, so you start to take the art out of the process and leverage more of science. It is that interaction of the, we call that the last mile, that is done through agentic. And so I think in composite form, with each contribution agentic makes in each module, it does actually develop the super underwriter.
Hmm. When I see your demo, Don, and I’ve seen you demo it for insurers and carriers, the ability for the Insurance Policy Suite to standardize prioritization based upon one dimension’s profitability, the other side is propensity to buying, right? And I watch insurers’ eyes light up. I look at insurance underwriter leaders looking at that and saying, wow, that is just terrific to see that 3×3 or the 9-box grid. It just changes how things go in that X, Y axis. Many, many carriers, there’s still so much reliance on static risk scores and backward-looking models. Why do you think this nine box grid, this form of prioritization essentially is more powerful? And how does it change what an underwriter actually works on during the day based upon your experience?
Tony, this is a lovely question. Thanks for bringing this up. So I’m the technical person, so I can very easily get lost in technicality. Now, everything I have talked about, the modules, there are two ways to look at this. You can either serve a piecemeal to an underwriter, so across multiple dimensions, there are multiple summaries for them to consider. And our solution does this. But what the solution also does is it feeds all of those inputs into a nine box grid. So it gives a summary. And that summary talks about here is your predicted profitability, but also it looks at behaviours, historical bind, win ratios with the brokers, and it also gives you a perspective on how likely you are to bind this. So ultimately that nine box grid is a starting point and we expect that an underwriter will use that, drill into it, have a look at what sits behind it, but more importantly look at the rationale. So every recommendation is backed up by rationale that’s been summarized and over a period of time the underwriters grow more confident because they see the transparency and they see the logic. Now why is this different from historical models? It is because historical models, to your point, are static, whereas the benefit we’re trying to deliver is something that is dynamic, so literally as your portfolio changes because a portfolio’s nuance is daily, right, you have a large loss, there’s a catastrophe and the nature of the book instantly changes, well it gives underwriters the ability to understand the lay of the land in real time and react to it. It also gives business leaders the ability to go into the platform and alter their appetite and the very second they alter that appetite it influences what the platform recommends people write. And so it’s a very real-time view for daily tweaks and controls and what you eliminate is lag. One of the challenges of the industry is that there are lags. If they come up with a view, underwriters pursue that approach and then they review feedback and then alter the view, and so they deal with this as projects maybe three, four times every year. Underwriting organizations might alter process. With the platform you can alter those processes every day should you need to. So it accelerates an organization’s ability to respond to market changes in almost real time.
Don, one of the most fascinating parts of managing capital allocation in a portfolio is understanding attachment points on reinsurance for underwriters. It’s a really strategic set of decisions. How does the Insurance Policy Suite help the underwriter guide themselves through those attachment points as they’re doing the underwriting?
Tony, this is a fabulous question. So we have based our data source on exposure management information. So whether it’s what’s pushed into globally recognized platforms such as AERA Touchstone or any record the organization has of what was underwritten, primary layers, secondary layers, deductibles, attachment points across the portfolio. And this is also the role that agentic plays. So it categorizes and summarizes exposure by layers and by attachment points, because we recognize that these structures actually have an impact on what treaties are in place, treaty limits, and whether or not the underwriter needs to purchase a facultative reinsurance just based on the development of that portfolio. So it is a critical deliverable for us to demonstrate exposure by multiple metrics, including attachment points. So an underwriter is then able to make a judgment on what points to attach at and the behaviors and performances at different layers.
Don, it just speaks to the complexity of underwriting. That’s just one more layer of being able to present it in a data-driven, decision-driven way. This speaks to the ability for underwriters to do their jobs even more effectively along the way because they know exactly what risks they’re taking and they know how their portfolio is balancing along the way. I think one of the biggest challenges in the market is the inboxes these underwriters have and how many submissions they have. We would be remiss if we didn’t discuss this yet. Most carriers admit that only a fraction of submissions ever receive even meaningful underwriting attention. There are so many even unread emails. And Don was mentioning how responsive brokers are expecting their underwriters to be and how that changes the brand and the relationship between an underwriter and a carrier if a carrier or an underwriter can quickly respond in time. But oftentimes underwriters can’t respond because they just don’t have capacity to respond. How does agentic prioritization expand underwriting capacity without just pushing more risk through the system and increasing exposure loss?
I think what agentic prioritization does is it prevents the wrong risk from consuming the underwriter’s attention. And that is how capacity creation is done in a meaningful manner rather than just pushing it through. So let’s unwrap it a little bit. And I’m going to call it a misconception. So the key misconception is that more capability means more risk gets bound. That to me is a misconception. I think in reality, agentic prioritization expands capacity by reducing the wasted human judgment time and not by just accelerating approvals. So today’s underwriters are spending an enormous amount of time on submissions that were not even viable to begin with or did not match the appetite or the pricing constraints. There is no reason for even a second for an underwriter to be spent on such submissions. So agentic prioritization intervenes before the underwriter’s effort is applied and capacity increases because underwriters now stop reviewing the noise. And that is valuable. So now, if we were to talk about, you know, you asked about increasing loss exposures. I think the possibility of you push more and therefore you possibly are going to increase your loss exposure. I think agentic prioritization, if governed well, and it must be governed well, with confidence thresholds and kill switches. Low confidence decisions are never auto-advanced. Stop them in their tracks. Appetite violations are blocked upstream. You’re not going to go anywhere when that happens. And with good guardrails around drift detection, which continuously tightens the decision boundaries, while you’re pushing more risks through to the underwriter, you’re not going to increase the loss exposure. Your books are actually going to be profitable, if anything, or more profitable. So the agentic prioritization, I believe, is only going to help write better books of business. And the subtle but very critical shift is that in traditional models, the biggest question was, is this risk good or bad? Now, in agentic prioritization, the question is, does this submission deserve an underwriter’s time and why? And that fundamental reframing of the question expands capacity in a meaningful manner and loss ratios remain protected and underwriter decision quality improves.
Yeah, it’s all about ROAC at the end of the day, right? It’s all about how the return on the actual capital turns out. And it just changed the game from being fast and focusing all the time on just even loss, but focusing on return on capital. Exactly right. But, you know, you brought up a couple of terms there that were pretty technical, drift and some other types of things like that. And, you know, there’s a lot of buzzwords in our business right now. Agentic AI is one of those terms. And from an underwriting perspective, what makes these, I’d love to know, kind of get your perspective on what makes these agents truly agentic versus, you know, just one more set of business rules, analytics, workflow automation, just with a new label. There’s so much agentic washing. I’d love to hear kind of a little bit more about the Genpact solution and why this is so different.
Sure. So I think that’s the right skepticism that we must imbibe in our conversations and the distinction matters. The simple test that we use is, can it decide what to do next when the path is not predefined? At Genpact, we’ve built a six-layer autonomy framework for our agents. And each of our agents in the Policy Suite is tested against those layers of autonomy. An agentic solution to us is not a flip of a switch where everything is either fully autonomous or not. It’s not binary. It probably never will be. And every agent that is deployed in the Policy Suite has gone through this, firstly, the basic question of can it decide what to do next when the path is not predefined? To what extent is it able to autonomously make the decisions and take the right path? And then you apply the policies of, do you want it to go autonomous to that certain degree? That’s a different question altogether. Now, rules, analytics, and workflows all assume that the decision tree already exists. The agentic systems don’t, and that’s what differentiates them at just the base level. The three traits that we look for when we’re talking about agentic underwriting versus rebranded automation is intent-driven behavior. So next step execution, how autonomous is it? What can it do without predefined steps coming in next? So traditional systems will say, if this, then this, else this. But an underwriting agent operates against an objective. So optimize the risk selection within an appetite and XYZ confidence bounds. The logic in this situation emerges from the intent and not from a hard-coded step-by-step workflow. And that, Tony, is what changes between what was the AI washing or the agentic washing is the term you used. How do you differentiate that agentic washing? The second thing is context accumulation and memory. So most underwriting tools are stateless. Each submission is evaluated as one off. Now, a true agent retains the context of the submissions and brokers and geographies at the time. It learns which signals historically change the outcomes and therefore prioritizes and inquiry behavior dynamically changes. That is another second very important aspect of how the old automation that focused only on efficiency changes when you apply true agentic solutions that are not rule-based, that are driven by an objective, but learning from memory and context accumulation. Context and intent become extremely important in that situation. And lastly, the autonomous judgment delegation. Agentic systems don’t just automate the workflow, and we’ve said that multiple times. They decide who should decide. Now, this is a very interesting one. A lot of times we hear that it’s about what to decide next as a step, but I think we start where first is you decide who should decide and then you decide what to decide. This is where a lot of your orchestrator agents or MCPs or planner agents, whatever we call them, that increases or improves the degree of autonomy in the entire Policy Suite because you’re calling the right agent for the right job to make that action at that time. So rerouting the edge cases as conditions or signals that evolve over time is another test of seeing, is this just following hard-coded automation rules or is this truly agentic? So those are some three or four different layers at which we test our systems, at which the Policy Suite has been tested. And it is truly autonomous to the degree we allow it to be as a policy for that client. And every client has different policies.
Well, it’s important in a regulated industry that there be such guardrails along the way. But boy, it almost raises the specter of straight through processing and touch free processing, which in certain cases, there may be some role for that if there could be enough controls. But one of the things you’re emphasizing is human in the loop design and where agents are designed to stop and where agents are designed to process. Don, I would love to hear where the Insurance Policy Suite intentionally stops agents from acting, and why is that boundary essential for underwriting accountability, regulatory compliance, and risk governance?
Tony, this is a fantastic question. Underwriting itself, even just the pre-bind series of processes, it entails more than a thousand decisions, and we call them micro-underwriting decisions. And so we’ve broken them into what is admin process, leading and naturally logical and seamless and anybody would arrive at that conclusion, and agentic is allowed to do this. And where do you get to a crossroads, and no answer is wrong but you need to make a decision and whatever decision you make will send you down a different path, that point is a human decision. And it goes back to the narrative around for an underwriter to actually earn their salt, there’s a lot of coming through they need to do and there are little judgments they need to make before they actually get to the point of underwriting, and so agentic is trying to do the heavy lifting. So in every single module, particularly the first module of the submission clearance, this is really task based, it is validating the identity of a submission, performing sanctions checks, performing regulatory checks, all of this can be executed by an agent with no human involved. And right at the end a file is prepared and sent to an underwriter saying here’s what I have done, I being the agent, here’s what I have done, do you wish to proceed with this? So actually at that point the decision to actually assess that risk sits with the underwriter. Now right through risk assessment what our platform does is it churns data, analyzes data, returns findings, makes recommendations, tracks the portfolio, points out trends, looks at a particular opportunity and points out any underwriting gaps that it thinks an underwriter should be aware of, right, this is thought leadership. And then it will make a recommendation but it will never move on to the next underwriting stage, so it is not autonomous underwriting. This is commercial and specialty lines, so it wants the underwriter right there in the middle, and what it is actually doing is just, think about this as for each underwriter, give that underwriter seven other underwriters just to look at the one risk and do all of that in real time. So each every stage of the process there are reports being presented in front of the underwriter saying it would take you three days to do this analysis and you might not do it as thoroughly as this, so here it is, I’ve just done it for you, what do you want to do next? That’s how the underwriter just comes through the different stages. We’ve given examples of what limits should you offer, what should be your attachment point, what coverage should you offer. Agent will suggest this and also clarify why it is suggesting it, but it leaves it to the underwriter to offer what coverage the underwriter wants. What is therefore important about this, if the underwriter uses the platform, it’s actually less about whether the underwriter accepts the recommendations. It’s more about if the underwriter rejects them, because there is the memory. And so the underwriter can circle back to this two or three years later to understand whether his deviation or her deviation was the right thing to do. So we’re very clear. We are not replacing underwriters. We are enabling them. We are empowering them. We don’t have a license to act as underwriting organizations. And so all decision points are for the underwriter to make. And a byproduct of all of this, but a very powerful byproduct, is the automation that happens behind the scenes. And so whilst we are maximizing the value of data, and whilst we prioritize the underwriter’s ability to make the right judgment, very informed judgments supported by internal and external third-party data, we are also mindful that there are multiple systems in every organization. For certain insurers per line of business they use 10 platforms and 10 underwriting systems, and so our platform actually tries to do the heavy lifting. It tries to bypass the need for a human to update multiple underlying systems or extract information from it. Our platform will do this. It’ll push information in. It’ll pull information out. It needs that information to support the underwriter making judgment. And when the underwriter has made a judgment, there will be a series of actions that need to be performed. And the agentic solution can prime that action and push that into underlying systems. And so whilst they’re getting the value of decisions-based and insightful data, it’s also solving for process automation in one fell swoop.
You describe it almost perfectly in terms of human plus AI. For AI, the agent in this case is almost an underwriting assistant. There were years where the submissions would be printed, stuffed into colored folders and passed around to underwriters where steps had to be done by underwriting assistants. And what you’re talking about here is how that gets done in seconds, probably isn’t the right answer, but in moments, getting it all pulled together and then providing a report, essentially, to allow underwriters to make the final decisions. And therefore, that’s the human plus AI specifics. You know, Don, one of the things you said, and I’ll ask this question to you if you would, one of Don’s earlier statements was around data, right? And, you know, you look at data in particular, there’s only so much that you get in a submission. And a lot of times, even when you get a submission, it isn’t exactly right, isn’t exactly complete. You can think about engineering reports and, you know, the wide variety of risks that are available out there for, you know, onshore energy or marine or specialty, you know, property areas, right? There’s just a lot of decisions that need to be made. There’s a lot of data to be used here. I would love to get a better sense. Since most of the underwriters only see what’s in the data historically that is provided, and adjacent hazards that are not disclosed in a submission could significantly alter a risk profile, the integration of external data into that decision-making process, we fumble, right? Where do they get it? How do they get it there? At what points do they provide it in that process? It sounds like your solution is doing this automagically almost. So when it’s actually provided to an underwriter, how is that shifting underwriting from being intuition-based judgment, like I think that there’s a fire hydrant down the street, to actually evidence-based selection where they’re actually seeing the distance between the location or fire hydrant or seeing the wildfire risk or the flood risks along the way?
Historically, underwriters have filled the information gap with intuition. And these are knowledgeable individuals. These are individuals that have years of experience and a lot of depth of knowledge, but it’s still intuition. And third-party data flips that equation by making what was invisible now visible, whether that’s adjacent hazards, secondary occupancies, crime density in a zip code. So many different things. And these are risks where underwriters always suspected, but couldn’t necessarily prove. And that’s why I call it intuition. And the gaps were filled by intuition. Now, most insurers assume the challenge is on onboarding of data feeds. While we believe that the real challenge is deciding when the data is a decision-changing data point. Like when you were framing your question, you said there’s so much data available. Sometimes just more data isn’t necessarily good data or doesn’t drive a better decision. You have to look for what data is going to have a material impact on my decision and therefore what am I consuming and how am I consuming? So if every data source, for example, adds latency, adds false positives, or is just simply overwhelming the underwriter, the intuition-based decisions will just creep back in very quickly. Change management will be a nightmare. And that is what has happened many times in the past. Now, evidence-based underwriting only works when the data is contextual. It is comparative. It’s actionable. And that’s what the Policy Suite does. So the agentic features decide which platform or which third party do I need to reach out to to get certain data elements in order to make that decision in that context for that particular peril that I can present to the underwriter. That’s what it does out of the box by connecting with external data sources. Say, Google APIs bring in latitude, longitude information of a strip mall, and suddenly the risk profile might change because I have a shop that’s lower risk profile, but adjacent to it is a laundry, which is a much higher risk profile. So those are things that aid decisions. It is contextual, and it is actionable. So I don’t need to just throw data points. And I think finally, I’d say that the third party data gives underwriters judgment, something that’s provable to stand on. And therefore, again, it goes back to what I said earlier about writing a profitable book rather than just going through more submissions and just writing more. It’s writing profitably.
I just simply have to ask because, you know, in insurance and, sorry, property heavily portfolios, underwriters traditionally have not all relied on complete full modeling of risks, perils, everything else. They’ve done this, used this intuition just because there wasn’t time and the brokers were clamoring for responses fairly quickly. But now you’re looking at underwriters who are actually making decisions with full modeling in the Genpact Insurance Policy Suite. How does your exposure management capability change what good enough underwriting looks like without slowing down the business? Because now there’s so much more in front of them and they have a full analysis. It would almost seem like they slow down maybe a tad a little bit because there’s so much more to ingest.
I think this is where exposure management stops being a modeling exercise and it actually becomes a real-time decision discipline. And our exposure management capability changes the question from, have we modeled this risk, to, do we have sufficient confidence that the risk won’t destabilize the portfolio? And that minor shift of asking that question. And if you think about it, earlier also I was talking about how you change the primary question for that activity, how you change and shift that, changes the behavior and changes the outcomes that are being driven. And this shift matters because instead of waiting for full models, the systems build new real-time exposure views using partial data, compare submission against current portfolio stress points, and flag accumulation, tail risk, amplification, and correlations. So most submissions almost never need a full model. So the way I see it, good enough no longer means a best guess under time pressure. It actually means decision-grade confidence at the moment of binding. So this will lead to reduced hidden accumulation and improved portfolio resilience. So that subtle but very critical shift of fundamentally changing what am I asking as the starting point of the question gets us to a better outcome. And that was the principle that we’ve imbibed in building the Policy Suite.
What you’re talking about is so transformational, Yasir. When you think about it, I think about the change management process in dealing with an insurer. Every one of these insurers had dealt with the five-year, seven-year journey of implementing a core policy billing claims suite. They’re sitting on $100 million, $200 million of depreciation of that particular implementation. And not just the financial part of that, but the bruises, the bloody mess of implementing a very complicated solution out there. And then they look at the ROI on those old solutions and they are super skeptical. And so they’re going to be skeptical of yet another large-scale transformation. But I think your solution is a little different. It’s headless, it’s modular. How does your solution reduce the adoption risk while still providing meaningful underwriting impact, given the history of what’s gone on in the bowels of underwriting and the bind and post-bind process?
See, the Policy Suite was designed with the assumption that the core is not going away. We are not replacing the underwriting platforms. We inject decision intelligence where underwriting is happening. So that was the fundamental principle of how the Policy Suite was conceptualized and without creating another monolithic technology to go after it. So how does a modular headless architecture reduce the adoption risk? There are two primary vectors to look at it. So firstly, when you decouple the decisions from transactions. So in the Policy Suite, underwriting intelligence is delivered as headless decisions. What does that mean? What it means is that it is not embedded every time customized into the platform. And every time you just are stuck and getting deeper and deeper into this customization journey, which is stuck into a single platform. A headless architecture gets you away from that. So let’s say a client already has ServiceNow in their stack as a workflow. ServiceNow will continue to orchestrate the work. Guidewire is their core system of record. Guidewire will continue to be the core system of record for policy administration. Genpact’s Insurance Policy Suite operates as an agentic decision layer invoked via APIs that causes different agents to autonomously take the actions we’ve been talking about without having to do a multi-year re-platforming exercise. And we believe that is what drives benefits at scale. The second piece of headless modular architecture is that the capabilities, having been built in a modular fashion in four discrete modules, can be deployed individually, some of them or all of them. Now, depending on where the client’s priorities are, you can take the modules to say that, okay, let me take one module for one single line of business, prove the value and then expand. Or you can say that I’ve already been on a journey. I’ve already built some technology or partnered with, say, let’s say for submission ingestion, but I need the modules that are doing risk evaluation and exposure management. So we can plug and play into and meet the clients where they want us to meet them. That way, the headless modular architecture helps modernize underwriting one decision at a time without making multi-year undertakings for huge platform transformations. And the beauty of it being that they amplify the value of investments in technologies you have already made because we’re not looking to rip and replace. We’re looking to fit like a hand in glove and improve the outcomes and the ROI.
I think about, I reflect on the conversations, the numerous conversations I had with insurers over the last year. And it seems like there’s such a, here at HFS, we call it a velocity gap between those who are using AI and really leaning into it and those who are just kind of waiting to see how things are going to play out. And it’s, you know, and then at home, you’re using it to do recipes and all these other things that’s built into your phone. But it seems to me that if we fast forward three years, there’s going to be a widening gap between data focus and traditional insurers. And it’s just that it seems to me like our data is showing that that is starting to develop in the real world. So, Yasir, what will separate insurers that implemented AI from insurers that actually changed underwriting economics? That’s a big difference. And what decisions do leaders need to make now to be in that second group?
I think, Tony, it’s an interesting one because the world we live in, time compression, has been fascinating in the last few years. Just a few years ago, we would be asking ourselves the questions, five to 10 years out, what do you think? And now it’s two to three years out, what do you think, right? Because the time compression at the speed at which we are evolving as, I think as humanity, not just industry, is just fascinating. But that aside, I picked that three-year thing in your question, so I’ll quit there. But, you know, I think most insurers will be able to say that they implemented AI. Almost everybody will say they implemented it. But the question is how many will be able to actually demonstrate that what they implemented materially changed loss ratios, expense ratios, and the quality of the book that they have underwritten. So in my view, the insurers that truly changed the underwriting economics would have done two or three things differently. So firstly, they move AI upstream of decisions and not to the downstream of the process. What that means is that AI does not sit behind workflows, but it will actually decide. Those that are going to use AI only to speed existing steps will definitely see efficiency gains. So now you have to decide whether you’re going to ring fence and restrict AI just to drive efficiency gains or leverage better economic value by letting it make the decisions. Of course, with guardrails, of course, with controls. The other thing is when you re-architect underwriting around confidence and not necessarily rules. So switching from AI-driven confidence scores that drive decision-making through the machine versus through a human intervention is going to be a big differentiator. I believe that will change the game. Data-led insurers will operate on confidence bands, dynamically adjusting the escalation plans. And I think we’re starting to see this and it’s starting to resonate really well with many business owners of using decision bands for escalation paths. And then you have most AI implementations stop at deployment. I think economic leaders will track which data elements change the outcomes, evolve the points of interventions. So constant reinvention is going to have to become a part of the DNA in which we operate. So the ones that are going to get economic leverage are the ones that are going to actually imbibe that and bring that into their DNA. So underwriting becomes self-improving, not just automated, not just a better and quicker system. So the winners will not be the ones that have most AI. I think they’ll be the ones that let AI change how the work happens, not just speed. So three years out, you and I might be here talking about, hey, what next? Are we going to teleport somebody from one place to another? And how do you ensure that? Who knows? Five years ago, we couldn’t have thought about some of the things that we’re living today.
Hard to believe. We used to say five years is fiction and three years is fact. Now it’s hard to say even one year is fact. You know, it’s happening so quickly. But I want to make clear to everyone who’s been listening to this conversation, I want to really thank Yasir and Don for joining today, is that the next phase of underwriting isn’t about adding more tools. It’s about changing how decisions are made. And that’s really what we’ve been discussing today. The insurers that are going to win won’t be the ones that simply implement AI. They’re the ones that are going to use it to focus on underwriting judgment, on the right risks, applying pricing discipline consistently, and they’re going to free their underwriters to work on what actually drives profitability and return on adjusted capital. That’s the difference between automating underwriting and transforming underwriting economics. And it’s the line separating tomorrow’s leaders from those that are still modernizing yesterday’s processes. So thank you all for joining today. I appreciate it.
Great. Thank you, Tony. Thank you, Don. Thank you very much. Thank you all.
Your account has been created. You can continue exploring free AI insights while you verify your email. Please check your inbox for the verification link to activate full access.
With the exception of our Horizons reports, most of our research is available for free on our website. Sign up for a free account and start realizing the power of insights now.
Our premium subscription gives enterprise clients access to our complete library of proprietary research, direct access to our industry analysts, and other benefits.
Contact us at [email protected] for more information on premium access.
If you are looking for help getting in touch with someone from HFS, please click the chat button to the bottom right of your screen to start a conversation with a member of our team.