Canadian Underwriter

Explain how AI leads to underwriting decisions, report urges

March 31, 2021   by Greg Meckbach

Print this page Share

Insurers who use big data and artificial intelligence to make underwriting decisions must be prepared to explain to clients and regulators exactly how they arrived at those decisions, an Insurance Institute of Canada researcher suggests.

“A challenge in the application of machine learning and artificial intelligence in insurance involves explainability,” senior researcher Paul Kovacs wrote in AI and Big Data: Implications for the Insurance Industry in Canada, a paper commissioned by The Insurance Institute of Canada. “Systems that learn to better anticipate outcomes of interest to the insurance industry must also be capable of explaining how they arrived at those conclusions.”

Artificial intelligence involves the use of computer algorithms to adapt their learning as they get new data. With AI, information technology systems can act on what they have learned to solve problems, reach conclusions, and predict behaviour, suggested Kovacs.

Some carriers are trying to get as much high-quality data as they can about consumers, Kovacs reported. This data can be used to better assess risk for new and existing consumers, and therefore determine what each customer should pay to properly cover their risks.

This could be bad news for some clients because, in some instances, it might lead to higher prices and reduced coverage options.

“Analysts may struggle to explain why a factor, or combination of factors, appears to have a significant impact on the assessment of risk,” Kovacs observed in the paper, released Mar. 22. “Consumers deserve to be informed about changes in their expected risk of loss, as reflected in the price they must pay for coverage, and it is important that this can be resolved if insurers are to rely on the emerging tools.”

One consequence of not helping the consumer to better understand the use of AI could well be that the insurer loses the business, warned Kovacs, who is also founder of the Institute for Catastrophic Loss Reduction and past CEO of the Property and Casualty Insurance Corporation.

“Some consumers will likely experience higher prices, reduced availability, and other unwelcome changes,” he wrote. “Companies unable or unwilling to prepare for these challenges will find that consumers are ready to move their business elsewhere, which might not be as simple as it was in the past, and will speak up about their treatment, which is much easier with social media.”

The report has eight recommendations for the industry.

One is to be prepared for uncertain regulation. Regulators will closely monitor the impact of AI and big data analytics, and insurers will need to explain why the new systems result in changes in the treatment of consumers, Kovacs predicted.

Another recommendation is to inform consumers and be proactive in demonstrating that application of the new algorithms are statistically fair and unbiased.

The industry should also accept “different views of fairness.” For example, people with a perspective on fair treatment of consumers may differ from people who have “an unfettered commitment to actuarial fairness.”

Insurers need to demonstrate an understanding of the growing importance of environmental, equity and social issues for consumers, embrace innovation, invest in new technology and to not be afraid to fail at first.

Kovacs also recommends insurers use big data to help create new products.

“Big data analytics can support the introduction of coverage for risks that were previously not insurable, like residential flood insurance,” he wrote. “Demand can be created for new products in situations where consumers may be unaware of the risks they face. There is scope, for example, for the insurance industry to adapt coverages provided to large businesses and homeowners to better serve small businesses and tenants and to expand cyber insurance coverages.”


Feature image courtesy of

Print this page Share

Have your say:

Your email address will not be published. Required fields are marked *