Takeaways:
- Missing nuances in cyber risk coverage can get 'expensive.'
- Only human underwriters can catch all the nuances in complicated commercial coverage.
- Insurers dedicating staff to AI need to mind differences in types of decision-making.
AI advances require commercial and specialty insurers to tread carefully on automation, especially for cyber risks, according to executives from insurers and an AI insurtech who spoke in a
Increased sophistication in
"I see a rethinking of how security architecture is built as we go forward, as AI becomes more sophisticated, as it is used offensively," he said. "There's going to be a rethinking of how simply we do business technology. A lot of new angles and new things are happening, in terms of how we defend businesses and protect their cyber assets."
In specific types of commercial insurance, like errors-and-omissions coverage and cyber-risk insurance, automating underwriting and missing certain subtleties can get very expensive, Tifft said. This means checking the accuracy of the information and catching characteristics of policies that human underwriters should decide, he added.
"Fundamentally, it's still a people-driven process in the specialty line space," Tifft said. "There's just too much subtlety in judgment. The risks that we're facing are so dynamic that it's hard to program automation to ingest and make decisions on those very dynamic market risks."
People will still have value in underwriting because of nuances and relationships AI does not have, said Rachel Switchenko, vice president of customer solutions and underwriting at Plymouth Rock.
"I think about how well the underwriters know their book of business and the things that data can't tell us," she said. "A data model or an AI model is going to tell them that this is the right decision — they'll still know differently."
Insurers' governance of their processes is changing, according to Nemanja Jokic, chief technology officer at Bound AI, an insurtech providing AI solutions to the industry.
"It's not just about software automation and deterministic system behavior," he said. "We've seen carriers and MGA partners introducing AI committees overseeing full governance [including] decision ownership, audit trails, explainability and escalation paths. Governance looks completely different depending on the decision type. What happens to the model if it fails at this specific task?"
In errors and omissions, for instance, Joki said, if an aspect of risk is missed or incorrect and then a claim is filed, insurers won't be concerned about efficiency problems.
"Governance is a huge part of it," he added. "We, as insurtech companies and carriers, are going through these evolutions. New things are popping up, so we have to be on our toes and ready for everything that comes our way."
