TORONTO — Manulife is using AI to review life insurance applications, speeding up sales and freeing up human agents to take on more complicated cases.

The Toronto-headquartered financial services giant began rolling out a new version of the AI system to its client-facing advisors last fall. The model is trained on historical data about how human agents have assessed applicants based on their medical history, age and other factors. 

The AI system—known as the Manulife Automated Underwriting Decision Engine, or MAUDE for short—can approve insurance applications, but it can’t deny coverage. Manulife categorizes policies by the risk of having to pay out on them during their term. For the firm’s “healthiest, low-risk applicants,” MAUDE says yes “within a matter of minutes,” said Karen Cutler, chief underwriter for Manulife’s Canadian division. It typically takes staff and automated systems another couple of days to generate all the paperwork.

If the model decides an applicant is higher-risk, it forwards the case to a human agent for review. “We want to find a way to insure as many people as we possibly can,” Cutler said. “We don’t believe that a model is going to turn over every rock to find a way to approve that case.”

Switching to AI review has also let Manulife simplify its application process, Cutler said. The firm’s forms previously asked people to enter a lot of information—a full medical history, for example. On Thursday, Manulife announced it’s replacing those forms with a system that starts with the medications an applicant is taking, then dynamically lists linked conditions so they can just check the right boxes. 

“Rather than the broad-brush, ‘Have you ever had anything wrong with you?’ it’s ‘What are you under treatment for?’” Cutler said. The data is then fed into MAUDE, which can figure out how each answer affects the risk profile of the applicant.

Manulife began exploring building AI models for underwriting a decade ago to replace more time-consuming and inflexible processes, Cutler said. Before the firm launched an early version of MAUDE in 2018, “an underwriter would review each and every application that came through the door,” Cutler said. Now, MAUDE does the first pass. It currently approves about 58 per cent of eligible applications automatically, and has signed off on tens of thousands to date.

Insurance companies have long used statistical modelling and other mathematical techniques to help price and make decisions about coverage. MAUDE is a significant upgrade on the rules-engine system that Manulife had before, according to Cutler. The older systems were “binary, very black and white,” she said. 

The rules engine had to be taught to say “yes” or “no” based on each factor—say a specific medication, health condition or age threshold—individually. Modern AI models can infer the right decision based on all the data available to them. They’re also easier to update when, for example, patients with diabetes switch medication en masse from Metformin to Ozempic, Cutler said.

Regulators and consumer rights groups have expressed significant concerns about how financial services firms might use AI tools to make biased decisions against customers based on historical data tainted by human discrimination. It’s a particular issue in life and health insurance, since sickness and death rates that vary by gender and race can be the result of societal and economic factors, not biological ones.

Cutler said Manulife tests its AI underwriting systems for bias. A team of underwriters and data scientists, as well as pricing and risk-management specialists, audits every new model for a period of time after launch, checking to ensure the decisions it makes are correct. 

Manulife also has to account for the risk of false positives, applications that MAUDE misjudges and approves when it shouldn’t have. The costs of those mistakes are priced into the firm’s policies, Cutler said. “Our human underwriters are not expected to be 100 per cent perfect either.” 

Still, MAUDE is helping Manulife sign up clients faster, letting it review more policies than it could before and boost sales. The firm has set a target to generate $1 billion of enterprise value using AI by 2027, a fifth of that through efficiency gains. Manulife ranks fifth globally on the Evident AI Index for insurance companies’ use of AI, which scores firms based on factors like research, talent development and responsible AI activities.

In addition to its underwriting models, the firm has rolled out an in-house AI copilot to all its employees. It also has 70 other AI applications for specific tasks like summarizing conversations in call centres, generating sales emails or assessing the security standards of potential software suppliers. 

Cutler said Manulife plans to give MAUDE more work in the future, with its next task set to be reviewing applications for critical illness insurance.