Over 25 years ago, it was difficult for the average person to imagine all of the things the internet could be used for. Today many, if not most people are in the same position thinking about the application of generative artificial intelligence (AI).
At scale, mass adoption is not something that’s happened yet, according to experts gathered for a panel discussion hosted by ratings agency AM Best, but many are close to it, says Arun Balakrishnan, chairman and CEO of Xceedance. His comments and others were part of a recent webinar, How Generative AI and ChatGPT are Reshaping the Insurance Industry.
Despite concerns, the promise of AI is huge, with companies exhibiting different levels of excitement, exuberance and caution, Balakrishnan adds. Experimentation is widespread and many “sandbox” environments – secure environments where programs are only able to access certain files and programs – are being created to evaluate the industry’s growing number of use cases seeking an opportunity to be tested.
“As you think about adapting the technology or deploying the technology, look closely in terms of what use cases make the right business value and the right business sense,” says Praveen Reddy, chief operating officer with Velocity Risk Underwriters. “It’s easy to get carried away but think closely and carefully around the right deployment of the technology.”
Alongside this, the experts say there is a real and pressing need for education, top-down messaging, to head off comparisons that a company’s own employees might be making to developments that they perceive are taking place elsewhere.
To arm those gathered for the virtual presentation with material for consideration when crafting those top-down policies and messages, the panel generously discussed data analytics, machine learning and technological developments preceding the advent of AI. They discussed how these developments are similar to and also quite different from those developments occurring today.
Use cases for generative AI, right choices and how companies are managing their wide range of development options were also discussed, as was product development for the world at large, which is generally going through many of the same changes.
Past, present and future
Actuaries have been doing data analytics forever, Balakrishnan says. The difference today is that their efforts are more computational now where they might’ve been more mathematical in the past. Machine learning, where software is coded to self-learn and improve with time, is not new either. The biggest limitation to machine learning adoption, he says, was and is the fact that those with data didn’t want to share it, while those with the technology didn’t have access to the industry’s data.
Things that did come of this era of development, he says, were more snippets in the form of robotics and automation.
“Generative AI took it several leaps forward. Here was a large language model which has already been trained to understand the nuances of communication, understand the nuances of situations, emotions and it’s been trained on a whole volume of data,” he adds.
A step back in time to six years ago, meanwhile, says Balakrishnan, and the talk was about how blockchain could make an impact on the industry. “But it required a network effect. It required various market participants to come to agree to working together. I don’t feel there is a limitation like this in adopting large language models,” he says.
What AI adoption is more akin to, the experts say, is the effort it took to move from mainframes to client servers to the internet and then to the cloud. “There’s always fear of data leakage and privacy not being secured and the risk involved. And regulation plays a big part, right? Usually when something blows up, regulation rushes in and maybe the pendulum swings to overregulation. I think the evolution here will be very similar,” says Ray Mirza, senior vice president with Berkshire Hathaway Specialty Insurance. “This is here to stay. It’ll evolve like every other technology shift that we’ve seen in the last 40 or 50 years.”
Present day, the panel referenced the experiments currently underway. While there is a lot of activity in the proof-of-concept world, says Reddy, in terms of real deployment, he says the industry’s efforts are in very nascent stages. Going forward, they say adoption of AI will allow smaller players to compete and larger players to differentiate. They warn, however, that regulators will also be paying attention. “We’ve got to assume that the regulators are looking at this and wondering how they can get their arms around it because it’s evolving so fast,” says moderator, John Weber, senior associate editor with AM Best.
Use cases, right choices and challenges
In personal lines, the panelists say there is a lot of experimentation occurring with customer experience applications. In commercial, there is more of a spectrum with small commercial insurers trying to use AI for underwriting, while larger commercial companies are exploring use cases for property risk inspection, says Balakrishnan.
At Berkshire, the firm’s first use case is related to submission processing. “That’ll be a two or three month effort that we’re just kicking off,” Mirza says. “There’s a whole use case pipeline flowing from there.”
Operational efficiencies are being looked at in at least half a dozen use cases at Velocity.
In one example, Reddy says the technology is analyzing accounts up for renewal and how they compare against the company’s current risk appetite. Those which do match are automatically pushed to an underwriter with a higher ratio of binding.
Risk evaluation is also being used to route business to the right underwriters – efforts that are in the earliest stages of development – “pre-proof-of-concept,” says Reddy.
Balakrishnan meanwhile, says agencies are also starting to use AI to compare binder language in the hope that it will reduce their cost base and error rates. “The effect could create very profitable agency models,” he adds.
Companies too will need to make choices – between using large language models which stay up to date about the open world, and those which are private and unable to access the outside.
“If folks are concerned about privacy or security, you make the right choice or the appropriate choice for that,” says Reddy who adds that such choices will depend on the line of business and use cases being considered. He also suggests that there may be more flexibility for AI’s use in commercial lines than personal lines. “You have to make the right choices with regard to the public versus private options.”
Products and profitability
Notably for the industry, Balakrishnan also directed the conversation toward the opportunity emerging for new products, as well: As other industries begin adopting the use of AI in their own work, lawyers, doctors, accountants and virtually anyone needing errors and omissions insurance, particularly those who will increasingly be relying on AI insights to make decisions, will soon find a coverage gap in their policies. “New coverages are going to get created,” he states. “It’s going to be hard for regulators and it will be hard for insurers. How do you price this? There’s no precedent in history. But there is going to be a coverage gap that I agree, we as an industry will need to solve for it.”
That said, he points out that cyber wasn’t a coverage until very recently. “The good thing about the industry is we’ve evolved over centuries now to cater to (needs and provide) the products of the times.”
