ANZ bank's insurance arm OnePath Life Insurance today revealed some early wins for its project with researchers at the University of Technology Sydney (UTS) which uses machine learning models and advanced analytics to improve the insurance underwriting process.
Unveiled back in January, OnePath's project with the university’s Advanced Analytics Institute (AAi) is investigating how client behaviour modelling, text mining and natural language processing, as well as social and predictive analytics, can “add value in the insurance sector”
As a result of the work, which has been ongoing for around 18 months, OnePath has been able to significantly reduce the number of questions it asks a particular customer segment when they apply for a policy.
"I've been underwriting for about 25 years, and in all that time we haven't really changed the way we do it. We still ask the same questions, we're not very good at real deep data analytics. And today's society has expectations, especially millennials, in terms of fulfilment: they don't want to wait," said OnePath's chief underwriter Peter Tilocca.
"Our goal was to utilise data and analytics to reduce the number of questions that we ask during the underwriting process and reduce the friction to improve the customer experience."
When buying most other forms of insurance, a customer would typically go online, answer some basic questions and get an instant quote.
"And away you go, it's really really simple. What we do in life insurance is a bit back to front," Tilocca said.
"We quote you a premium first, then we ask you to fill out a medical history and lengthy questionnaire, and then depending on your history we may come back and say 'actually your premium has increased, we have to exclude your back because you've had a back problem in the past'. And in a worst case scenario we can't accept to cover at all," he explained.
The result is a negative prospective customer experience, that may not even end in a policy.
"It's a lot of time and effort that's gone in to reaching that point. If it's rejected nobody wins in that scenario," Tilocca added.
UTS researchers, led by associate professor Guandong Xu, worked with 10 years' worth of data extracted from three core systems (some "tens of millions of rows of data") and applied advanced analytics and machine learning techniques to make connections between customer cohorts, the questions and answers, and resulting claims.
In a visualisation displayed at the CeBIT technology event in Sydney today, a heatmap grid showed where correlations between question responses and claims was high. The analysis enabled OnePath to pick out the most relevant questions, and discard the least relevant.
Late last year a successful pilot looked at the tradespeople customer segment, and was able to reduce a section of the application process from 32 questions to just seven without "any degradation to claims" Tilocca said.
"Essentially we're getting the same decisions but with a much better customer experience. With 10 years' worth of data, the risk is minimal. All of those questions don't add any value to this customer segment," he added.
The slimmed down questionnaire is now live.
Other analyses on the decade's worth of data have sought to identify which underwriters award claims incorrectly, or make exclusions that shouldn't have been made most often. It is helping OnePath, the life insurance division of which is being acquired by Zurich, establish improved quality assurance insight across its 55 underwriters.
Tilocca said that AI is improving a useful tool for underwriters, but could never replace them.
"Regardless of the technology we introduce and the automation, it's still a people business. A highly skilled underwriter is still worth their weight in gold, maintaining relationships and identifying the softer risks. Underwriting they always say is part art, part science. I'm just trying to increase the science," he said.
The AAi has worked with a number of financial institutions including Credit Suisse, Westpac, AMP and HCF. It has also worked with government agencies including the Australian Taxation Office and the New South Wales Ministry of Health.
In 2016 the institute worked with wealth management group Colonial First State to build advanced analytics models to observe complex relationship patterns between investors, employers and financial planners and build predictive models to help improve customer service.
“We aim to be the leading research group in applying data analytics and AI in FinTech across various wealth sectors such as insurance, superannuation and investment portfolios,” Xu said.