Paul Carroll
What's the current state of play with generative AI? It's been two years since ChatGPT launched, and while there's been a lot of publicity, I wonder how much are people actually using it versus just talking about it.
Fady Khayatt
Generative AI has certainly grabbed headlines regarding the possibilities in those two years. Over the last year or so, the insurance industry has been working to understand generative AI's potential, separating hype from reality, identifying use cases, and starting to test solutions. This has primarily involved relatively small-scale, cautious pilots.
There's been an awareness about not diving head-first into massive investments, with some tentativeness about what this technology will really deliver and what the return on investment will be. Some players have been burned by previous promises around technologies like blockchain, where there was a lot of hype, potential overinvestment, and no real delivery. This caution has meant that much of the experimentation and evolution has been more iterative than transformational.
Rather than true experimentation, the work with generative AI has been focused on deploying tools and copilots from providers into current processes, instead of considering how we can use this technology to transform our processes or make a step change in how we interact with customers, distributors, and internal stakeholders. This has somewhat limited the impact of these pilots and the identification of generative AI's real potential.
Paul Carroll
At this point, it seems we can divide the use of generative AI into two buckets in the insurance industry. One is becoming more efficient, and the other is actually using AI to make decisions. From what I've heard, the uses are more about AI gathering documents for agents, claims adjusters, and underwriters, rather than gaining insights for more effective underwriting. Does that match what you're seeing?
Fady Khayatt
The use cases we've identified span both categories. There is recognition that generative AI can improve both efficiency and decision-making. However, the latter is very hard and requires a more transformative change than just providing access to generative AI tools. It requires bigger changes to processes, governance, and structures.
The pilots that have been run and systems deployed have tended to focus more on the efficiency side. There's recognition across the market about the risks of this approach. At Oliver Wyman, we conducted a survey earlier this year of CEOs from companies listed on the New York Stock Exchange, and around 40% expressed concern about not moving fast enough.
Giving everyone access to tools like Microsoft Office Suites with Copilot, or other AI modules integrated within existing systems, has been helpful and driven significant usage, including internally at our organization. However, the tools are primarily making existing processes easier rather than changing how people make decisions or improving the quality of those decisions.
There's been much discussion about usage within underwriting and claims, particularly regarding changing the balance between art and science in complex underwriting and claims handling.
Paul Carroll
Are there any standout examples you've seen that others should try to emulate?
Fady Khayatt
In terms of efficiency, as we see both in our surveys and in our conversations, Gen AI has a huge impact on code and software development. That's probably the leading area of deployment and has driven a lot of efficiency. But those processes where it's been easy and quick to deploy are not really core to insurance industry priorities or needs.
We do see some examples of insurance companies deploying Gen AI into their claims process in terms of collecting unstructured data across handwritten claims files or hundreds of claims files that have been stored but are difficult for a claims adviser to go through. We are starting to see insurers using Gen AI to go through their archives and identify what critical items determined the outcome. We are starting to see some of that deployment, but it's relatively limited.
Paul Carroll
What do you recommend to folks you talk to about what they ought to be doing now and where they ought to be trying to get over the next year or two?
Fady Khayatt
I think there's a question about identifying where internal development of Gen AI is worth focusing on and worth investigating. There are areas where Gen AI will have a big impact but where it's not necessarily right for insurance companies to lead the charge. Gen AI is going to be very relevant to areas in the value chain around IT, HR, and legal and compliance. However, these aren't necessarily areas where insurers want to be developing their own proprietary solutions — there may be industry solutions they can deploy instead.
The key is identifying areas that will create a distinct competitive advantage if insurers take the lead. This will be different for different players depending on their areas of focus and strategic priorities.
The second point is ensuring alignment with broader transformation objectives within the business. What we've seen so far is some Gen AI experimentation that's disconnected from broader change programs. You'll get more traction by integrating Gen AI thinking into existing transformation goals, whether that's developing a new line of business around energy transition or cyber, or upgrading the underwriting workbench.
The third element is making sure that the focus is on both top line and bottom line. There's been a lot of focus on efficiency, but we need to understand these opportunities from both a growth perspective and an efficiency perspective. Otherwise, you're looking through too narrow a lens.
Finally, consider whether the capability you're going to deploy is specific to your organization. You need to assess whether to build it yourself, work with a partner, or adopt an industry standard solution. Not everything needs to be built in-house, and not everything should simply be taken off the shelf. There's a lot in between, and being open to all of those options is important.
Paul Carroll
You bring up an interesting point about efficiency. I've heard people discuss the importance of triage processes within insurers and whether AI can support this sort of decision making. How important is this in your view?
Fady Khayatt
I think AI can play an important role in triage. Often, when we talk about triage, it's about underwriting or claims triage in terms of simple cases versus complex cases. You could deploy generative AI to help with these processes, but beyond that there's potential for generative AI to also support case underwriters or claims handlers in dealing with complex cases and claims.
However, there are other parts of the value chain where AI can help with decision making, with similar benefits. For example, in the sales process, this depends on markets and products, but identifying the next best action or next best product for a certain customer is another opportunity. AI could also be deployed in retention processes — if a customer calls to cancel their policy, there are opportunities there.
Beyond that, there's generating content and personalization of marketing that's going to be sent to the client.
So I think you can see Gen AI being deployed across many different parts of the chain. For example, in the U.K., for a personal lines player, the retention process is critical and can be very differentiating in terms of outcomes. Choosing to invest there rather than in the front end of distribution makes sense in a market dominated by price comparison websites.
Depending on the market and line of business, there are opportunities in lots of places, but the critical opportunity for that business model in that market will be different.
Paul Carroll
Looking ahead, what are your thoughts on how things might change over the next few years, particularly regarding “agentic AI”? I'm personally concerned about AI systems being set up as agents that can take autonomous actions on our behalf.
Fady Khayatt
My kids often say I’m like “an old man shouting at the wind” because they think I underestimate the impact of these technologies. But in a heavily regulated market like insurance, there's going to be significant caution before implementing tools that execute multistep processes or make autonomous decisions. When you consider the level of regulation around ensuring fair customer treatment, and proper capital adequacy, deployment of fully autonomous systems will take time.
We've already seen this with generative AI chatbots and voice services. In Europe, particularly, we're very cautious about deploying anything that touches the customer because it's not just about reputational and brand risk — there's also regulatory risk to consider. I believe this cautious approach to deployment will continue within the insurance industry.
Paul Carroll
Makes sense. Anything else either in terms of the future or just anything I didn't ask you about that you wanted to touch on?
Fady Khayatt
I think the key point I'd touch on is how to move to a more transformative approach with generative AI rather than an incremental one. We've talked about aligning with broader strategy and being clear on impact areas that matter for business differentiation. But there's another crucial aspect we haven't discussed: actually making transformation happen. There needs to be a business-led change rather than a technology-led change. If generative AI is really going to fulfill its promise, it has to change how key people in the business work and fundamentally change those processes.
This transformation needs to start not from what generative AI can do or what's available to integrate into existing processes. Instead, it should begin with identifying where the current processes are broken. Can we reimagine how we'd like that process to work ideally? Once we've done that, we can think through whether generative AI can help us get there. There might not be anything off the shelf that allows us to achieve this, but the technology would enable us to get there.
Then you can work with IT teams and vendors to develop something that delivers this transformed process, rather than just taking existing tools and slotting them into our current process. The impetus has to come from the business or really engage the business, which I don't think happens enough today.
Paul Carroll
I just came from a conference where an old colleague of mine made this point on a panel with CEOs of big insurance companies. They were talking about what the technology could do, but Andrew said, "Forget the technology. This has to be driven by business people." They were taken aback, but it was a good point to make.
Fady Khayatt
It has to start with the business saying, "We could do this better." Whether that's customer interaction, underwriting, or financing, we could do these things better. We could see that if we could extract more from our data, improve our interactions, or change processes fundamentally, it would transform both how we present to the market and our operational efficiency. Now let's see if we can either co-construct or find solutions out there that allow us to do it. And I don't think there's enough of that thinking going on.
Paul Carroll
Thanks, Fady. This is great.
About Fady Khayatt
 | Fady Khayatt is a partner at Oliver Wyman in the European Insurance and Asset management business. Based in the London office, Fady has worked around the world for insurance businesses covering strategy, operations, governance, risk, digital transformation, and capital management. |
