The adoption of generative artificial intelligence (AI) like ChatGPT is projected to take off across the insurance landscape, with one expert putting the timeline at 12 to 18 months.
Vikas Bhalla (pictured), executive vice president and head of insurance at data analytics and digital solutions company EXL, said that most insurance companies will be exploring use cases for generative AI and large language models across a range of functions during that period.
But he cautioned that even as traction grows for AI, it’s extremely difficult to predict what its use will look like.
“What you will see over the next 12 to 18 months is a progression, as the technology becomes more recognised and more accepted,” Bhalla said.
“People will learn how to manage the risks associated with it, and insurance organisations will move from employee-facing to rep-facing to customer-facing uses of AI. You’ll see the impact really going up, and that is going to be a big change.”
In a Q2 2023 earnings call, the CEO told investors that applications of large language models would be iterative, and therefore take more time to produce benefits for insurance companies than “breathless rhetoric” in the industry implies.
Bhalla agreed that it’s too soon to see what form such technologies will take even as observers speak about AI’s increasing ubiquity.
“The form that such technologies will take six months to a year from now will be very different… because the pace at which new disruptive technologies is increasing,” Bhalla told Insurance Business. “It’s extremely difficult for one to predict what a form of that is going to be.”
Despite this, insurance companies are keen to deploy customer-facing AI solutions, according to Bhalla. EXL, which works with large insurers and brokers worldwide, said it has seen a “frenzy” of client interest in ChatGPT over the past few months.
According to EXL, the most popular initial applications for generative AI in financial services, including insurance, include:
However, there are hurdles for insurance companies to overcome before any significant generative AI usage takes off, EXL cautioned.
The company tells clients that data governance, data migration, and silo-breakdowns within an organisation are necessary to get a customer-facing project off the ground.
“Will insurers have tried [generative AI] in something [within 12 to 18 months]? I think yes,” Bhalla said.
“Would they have scaled it up significantly? In my view, that's going to take a bit more time. It will depend a lot on the learnings and constraints that we see. There's still a lot of regulatory approvals and changes needed before companies can scale up.”
Bhalla shared three recommendations for companies experimenting with generative AI: using closed data sets, keeping a human in the loop, and slowly progressing usage over time to minimise risk.
“When you look at creating of your first few implementations, the AI should be applied only to closed data sets,” he said. “You can take a pre-trained large language model, but you need to train it on your own data limits initially.”
Organisations should avoid combining their internal data with external ones, and refrain from exposing their data to the external, Bhalla advised.
“The second thing we telling clients is to have human in the loop,” the insurance head continued. “You can’t delegate the decision making and running of the operation [to AI], whether it is new business, underwriting, or claims. A human in the loop is important because you need to make sure that there is a checking mechanism.”
Finally, insurance companies can manage their risks by progressing the penetration of disruptive AI technology. Customer-facing AI applications are deemed the highest level of use, and therefore the riskiest.
“We recommend our insurance clients to start with the employee-facing work, then go to representative-facing work, and then proceed with customer-facing work,” said Bhalla.
Is your organisation exploring use cases of generative AI? Tell us about your experience in the comments below.