Boundaries need to be set before insurers start investing in AI and applying them across the insurance value chain, experts said during a webinar on Thursday (Sep 12).
The power of generative AI (genAI) to analyse vast amounts of data would mean underwriters can access applicant risk levels faster and in a more accurate way than traditional methods, said Jon Holvoet, Chief Technology Officer at Belgian credit insurer Credendo.
But boundaries will need to be set before large-language models (LLMs) are implemented in insurance, to ensure that sensitive data isn’t used to train external models — and that they would not be spit out to competitors or other actors through prompts.
“You have to make sure that your risk and your compliance and your legal department and your procurement department are all in the loop in from the beginning,” said Holvoet. “Before allowing anything to be set up, we’ve actually spent half of the time making sure that whatever we set up protects our boundaries, that none of the data is used to train any external models where we don’t have control.”
Holvoet said Credendo went through the fine print to ensure that the insurer had all the guarantees that it needs.
Emerging AI innovations are also introducing new risks for insurers, even as most of them claim to be on top of the problem, with most Asian insurers ‘clinging to traditional organisational structures’, according to a McKinsey report.
Notably, in a high-profile move, Samsung banned employees from using generative AI in May last year after it was discovered that sensitive code had been uploaded to the platform, citing concerns that data transmitted to AI platforms might be disclosed to other users.
Holvoet said that domain-specific LLMs can also enhance accuracy and reduce the risk of hallucinations. “It would be great if we saw insurers working more and more together on some of these common areas, like fraud detection, to create domain specific, fine-tuned open source models,” he added.
Still, humans will need to be in the loop, and models will have to be trained to cite sources. “AI is wonderful but it also makes mistakes,” he said. “On our implementations, you do see that talkative parrot syndrome that AI struggles with, so the human validation (is important).”
Apart from the risks and limitations LLMs have, humans will also have to learn how to write good prompts, Holvoet said. “It’s not only about the technology and giving your employees access to that technology, but there is a real need to train and educate the people as well,” he added. “It takes time. It takes practice, and you have to invest in that within your company.”
Hyper-personalisation and product management
Technology — previously used to automate processes — is now being used to enable users to change decision models based on real-time data, said SAP Fioneer’s Managing Director Chirag Shah.
Insurers are also designing “really complex product configurations” driven by AI, enabling users to make more informed decisions. “I think this is a fundamentally different approach,” Shah said. “When we talk about business efficiency, I think we are talking about making a lot of processes AI-enabled, AI-supported, in order to improve the efficiency of an entire process.”
AI can be incorporated into operational management, Shah said, allowing managers to make informed decisions through automatic data analysis for instance. It can also help in improving quality assurance and enabling partner ecosystems to deliver projects.
There’s also the aspect of hyper-personalisation that generative AI can help offer, said Madhu Malhotra, Chief Technology Officer at Indian GI insurer Zuno.
Zuno is building “a lot of personalisation, engagement and communication work” on the genAI front, with chatbots used for personalised, custom interactions, she said. “India is a vast market, and in order to reach all touchpoints, I think technology has a lot to do with that.”
But as customers start and explore on their own the policies that best fit them, AI driven hyper-personalisation will be a key differentiator.
“If you are doing personalised engagement and the products that are being recommended are suited to the customer’s needs, then I think that B2C channel is definitely going to grow, because that is what is untapped right now,” she said. “AI is going to power a lot of this personalised engagement and product recommendation.”
