Artificial intelligence is becoming a core competency for insurers that want to stay competitive in the face of geopolitical volatility, shifting claims patterns, persistent cost pressures and rising customer expectations. But to unlock AI’s full potential, carriers need to first navigate shifting workforce dynamics, modernize outdated legacy systems and overcome their tech-scoping problem, said panelists during a recent webinar from Federato.
Unlike other technology trends like cloud or mobile that really only transform one area, AI is “about fundamentally rethinking business models and creating new opportunities, and it affects everything across the value chain,” said Mark Breading, partner at ReSource Pro. He said that “the Internet of 30-plus years ago maybe is the closest parallel.”
AI can distribute expertise
The insurance industry is in “the early stages of a generational transition,” said Federato CEO and Co-founder Will Ross. Workforce dynamics are shifting as senior-level executives retire, causing a “turnover of decision-makers,” he said. This means newer employees are being asked to handle increasingly complex underwriting tasks they may not be ready for.
Karlyn Carnahan, head of Insurance, North America at Celent, agreed, noting that a lot of the basic work trainee underwriters started on has been automated, leaving more complex decision-making.
“If you have a new workforce coming in that doesn’t have the 20 years of expertise, there is this question of how do we assure that the right decisions are being made. This is where I think AI serves as a huge augmentor,” she said. AI can play a major supporting role by guiding newer underwriters — “but that AI then has to be built on the expertise of the organization,” she said.
“Some of the skill sets that we’re going to need will be quite different in the future,” Carnahan added. “We’re going to need people who understand how to design prompts in a way that what I call a ‘boss agent’ can break that up into a variety of steps, which means we still need people who understand the steps and the logic. And we will need a framework for managing the agents and for managing the governance.”
Instead of performing very repetitive tasks like entering data, she believes “people are going to be more supervising some of these AI-driven workflows. This is going to accelerate decision cycles dramatically because AI can adjust the data and run the simulations and autonomously do work like drafting policies.”
“One of the other aspects of this workflow is that the knowledge becomes more distributed,” she said, which “reduces the dependency on any individual person” and allows their experience and expertise to benefit everybody “because the AI can learn from that.”
AI can use that knowledge to help carriers avoid repeating historic cycles of losing underwriting discipline in soft markets, Carnahan said.
While insurers typically shed business during a hard market cycle, as the market softens, carriers tend to “get a little bit stupid and start bringing it all back in,” she said, even if they “don’t necessarily have the skill set or the data to be able to appropriately underwrite it.”
“We lose our discipline,” she said, “and we believe that we’re smarter than we are.”
AI may be able to help break that pattern, Carnahan said, because it can pull together massive quantities of data so underwriting “isn’t dependent on one human being’s knowledge of a very obscure part of the market.”
Legacy infrastructure is a barrier
Legacy systems pose a major barrier for carriers looking to leverage the full potential of AI, the panelists said.
The problem is that legacy systems are often data silos, Carnahan said, especially some of the older solutions. These systems “don’t necessarily talk to each other,” she said, which causes fragmentation, leading to duplication of effort and data.
“There’s just a lot of challenges with the legacy systems today,” she said. “Don’t get me wrong, they do the basic that they’re supposed to, exactly as they were designed. They manufacture policies; they manufacture claims checks; they produce bills exactly as necessary, but we need more than that today.”
“A policy admin system is very good at manufacturing a policy; it’s not good at underwriting,” she said. “There are still all these things that underwriters do outside of a system.” That makes it very difficult for others to access that knowledge and use it at the strategic level.
Carnahan also noted that because “these systems are hard to configure or code, changes take months and are often expensive.”
Legacy systems are a drain on resources, Breading agreed. “When you have half of the IT budget that’s consumed just to maintain these core systems, that in itself is a problem… It makes it difficult to get to market fast with changes or new products, and then it constrains your resources for other, more innovative projects that you want to pursue.”
Carriers have also made a habit of layering new systems on top of old ones, the panelists said.
“The older legacy systems took so long to get implemented that once you were in the process you didn’t want to derail that implementation,” Carnahan said. “We have these carriers that are burdened with many policy admin, billing and claim systems, and the cost to sunset them and to roll them all in is sometimes difficult to cost justify, and so they end up leaving them there.”
“That’s where we get to this issue of these rigid data models that are in silos and difficult to integrate, often with batch processes. Finding the way to integrate the workflows, to integrate the data, to integrate the overall process can be really challenging,” she said.
Breading noted that “most of these systems were designed based on the traditional mindset of the industry, being very conservative… [with] sequential processing and a static set of requirements.”
But “we’re in a very dynamic world,” he said, with access to real-time data and a fast-moving technology landscape. “So, the whole mindset of the design and how they were architected doesn’t really fit with today’s world.”
Breading said there’s been incremental modernization over the last 20 to 40 years where carriers are “trying to bolt-on and adapt some of the newer technologies, the newer capabilities into this traditional core.”
The problem is that trying to bolt an AI chat interface onto legacy framework doesn’t really work because the underlying technology is too slow and fragmented, Ross said.
You can tell the chatbot to go rate a policy with certain criteria, he said, but “if the product definition of the underlying core system was built on a certain era of technology, the time it may take to get that rate back will be depending on 14 different API calls to get to that rate and a seven-minute time frame to turn that around.” That isn’t practical when people are used to instant results from a Google search.
Carriers have a tech-scoping problem
Carnahan said she’s seen tremendous success in AI, but she’s noticed that carriers often stumble when it comes to scope. Some carriers are thinking too small, she said, “using AI on very limited aspects, and so they’re not getting the kind of returns that you would expect them to get. Or the opposite, they’re thinking so big that it becomes unwieldy and almost impossible to govern or measure.”
Neither one of those approaches works, she said. Instead, “carriers need to be thinking very carefully about what it is they’re actually trying to achieve and make those use cases big enough to matter but not so big that that they fail because they’re trying to boil the ocean.”
Ross agreed that there’s a tech-scoping problem. He said he’s been pushing customers to scope their projects more toward the middle, not just by line of business, so that you have an answer when senior leaders start asking about the bottom-line impact for the organization.
“You want to have an answer for, ‘what’s the next corner I’m going to turn capability wise, and then what is the scale of impact I’ve had’—and it needs to be big enough that they can see the next phase and the next phase,” he said.
Near-term priorities
Looking ahead, carriers need to focus on modernizing product definition to speed quoting, Ross said. If the bind-and-issuance process still takes a while, that isn’t really a problem, he said, but the quoting process needs to happen fast. That means you need access to your rating and forms attachment logic.
“If that part of the product definition is living in a legacy enough architecture that you can’t get to that speed outcome for the quote, then I think you’re in a world of hurt,” he said.
Carnahan agreed that product definition needs to be housed in a more modern way. She also stressed the importance of understanding workflow and processes. “If you don’t understand in the first place how you do your work, it’s difficult to have a vision of where you’re going to be applying the AI in the future,” she said.
Topics
Trends
InsurTech
Data Driven
Artificial Intelligence
