Underwriting, insurance executives say, is more art than science. But artificial intelligence could reverse that relationship.
Insurance underwriters determine the risk involved in insuring people or assets – and how much a customer must pay them to assume that risk.
“Underwriting is an art, not a formula,” said Calvin Lim, co-founder and CEO of UW Insure Brokers, a Canadian specialist in property & casualty brokering, speaking at the recent Insurtech Insights conference in Hong Kong.
Nonetheless his firm has launched BrokerAid, an AI product to help insurers such as Aviva evaluate car insurance policies to detect fraud. “AI is crucial to controlling loss ratios” on products like motor insurance that often don’t make money for the insurer.
US ahead, Asia catching up
Underwriting might still be a human endeavor, but Lim says AI is good at spotting patterns to make it more effective.
North American insurers have been early adopters of AI within the underwriting process.
Henry Wong, vice president of business development for Asia at Munch Re, says there are 11 companies in the US “heavily using AI for underwriting”. The reinsurer has developed its own AI tools; he says two carriers in Southeast Asia are using it.
“We’re using it to push the frontier of efficiency and to improve decisioning abilities, not just for simple tasks,” Wong said. “This is no longer about proofs of concept.”
However, uptake in Asia has been slow.
Wong says US insurers have access to more data, and to better-quality data. “In Asia, the data-readiness is not there,” he said. “Companies struggle to marry claims data to underwriting data to make a decision at point of sale.”
Extract and predict
But insurers in Asia are finding areas where AI is useful.
Rebecca Zhang, head of regional partnerships at SCOR Re, says insurers are using AI to improve operational efficiency, such as by using internal AI agents to extract and summarize information from unstructured documents, such as lengthy medical records.
Now insurers are experimenting with generative AI to improve their underwriting models’ predictive capacity. “There are a lot of applications for this in underwriting,” she said, including cover for critical illness.
“As a reinsurer, we see risk capacity is shrinking for comprehensive critical-illness products,” she said. “Tools that can control risk will help,” such as those that can parse a customer base to determine who’s a higher risk. For example, SCOR’s software helped a carrier in China work out that a claimant in a Beijing hospital had already been hospitalized in another city with a preexisting condition.
HSBC Insurance has been using AI tools for such things for several years, says Karina Au, chief underwriting officer. “We can rely on straight-through processing in the mass market for whole-life plans, savings plans, medical and critical illness,” she said.
She regards AI as necessary to support scaling the business, by enabling human underwriters to quickly search documents and do more business, as well as helping with customer service.
Not ready for prime time
These comments were made on stage, where there’s an inclination toward bullishness.
Privately, insurance executives are more cautious. They see the possibilities of AI, particularly large-language models, to support a huge scaling of business.
“GenAi is a force multiplier because it creates, on top of what traditional AI can do,” said one insurer’s head of venture investment. This includes clearing away a lot of the admin and paperwork that consumers an agent’s time, and better connecting customer service reps to corporate and product details.
But insurers aren’t ready to unleash genAI directly on customers or sales agents. “Risk and hallucinations are barriers to scale,” the investment executive said.
He noted that an LLM that is correct 99 percent of the time is still an unacceptable risk for underwriting or customer advice: that would mean an insurer with a million customers is asking for 10,000 complaints or lawsuits, which could cripple its reputation.
And RoI
GenAI is also expensive. Large enterprises are keen to build their own small-language models that are restricted to analyzing proprietary data. This improves accuracy and safeguards IP and customer data, but it’s costly to build and maintain. Insurers are better off partnering with fintechs to help with a particular application, and avoid making pricy investments into AI infrastructure. “We need insurance-specific use cases that don’t require huge compute,” he said.
An insurance CEO says it’s one thing to use LLMs for accessing information or helping call centers, and another to scale these into revenue-generating processes, such as for agents.
“With scale comes cost,” he said. “You’re now going to the cloud, using a lot of micro-services, and you face a new level of security and compliance risk. Our CFO looks at this as a lot of new costs without a return.”
He added: “The solution isn’t with tech companies: it’s about our organization. Yes we’re seeing internal use cases, but we haven’t seen a breakthrough use case that doesn’t come with big risks.”
Back to data
The biggest challenges, however, go back to accessing quality data.
One reinsurance CEO says AI tools need a lot of granular data to populate dozens of fields for a given application. She says the industry in Hong Kong talked about forming a consortium to pool investments in data, because the costs are too high for most carriers to assume alone.
That initiative went nowhere, however. One reason is that insurers fear that if they all use the same data, they can’t distinguish themselves. The industry is more likely to look to proprietary data as firms seek a competitive edge.
Another challenge is explainability. “Carriers that can’t explain why their AI made a decision will be challenged by their stakeholders,” said Munch Re’s Wong.
Business structures in Asia may also be holding back use of AI in underwriting. Lim from UW says many general insurers in Asia focus more on generating high premiums and relying on the investment returns than on careful underwriting. “That’s not sustainable,” he said.
For insurers of all types and sizes, though, AI is going to be an important tool of upgrading their underwriting skills, which in turn should enable them to expand existing markets and create new ones. The speed with which this happens correlates on their access to data more than tapping an AI company’s algorithms.
