Australian insurers have been busy adding clauses to their privacy policies to inform customers their data has been fed to AI models.

In late March, NobleOak added a disclosure in its privacy policy that its AI models “ingest personal information”, joining Australian Unity, Rest, HESTA, Pet Circle, Resolution Life Australasia and the local branch of AXA XL, which made similar updates to their policies between now and late last year.

The policies do not list the types of personal information that the AI models process or are trained on, and provide no specific use-cases; except for general insurer QBE Australia, which cites “machine learning tools used to detect the risk that a claim may be denied” as an example.

Several other insurers that have not added AI-related disclosures to their privacy statements have still detailed how they have piloted and deployed AI across backend operations and customer-facing services, largely to make them less time-consuming and more personalised respectively.

Suncorp told Information Age that it has “deployed hundreds of AI models” and cited examples across “sales and service, claims and supply chain management”.

Insurance Australia Group (IAG), QBE Australia and MLC Life Insurance, which declined to comment, and Resolution Life Australasia, which did not reply, have all built AI tools for high-stakes tasks like partially automating claims assessment.

AI – in combination with access to non-traditional and more granular data feeds like aerial imagery, daily transactions and smart cars – is also enabling the sector to move from traditional risk pooling to assessing consumers and pricing products at a more individual-level.

AI-enabled pricing

“Actuarial models for risk, cost and demand modelling” are among the insurance sector’s “most common uses” for AI, according to a report the Australian Securities and Investment Commission released in October.

The watchdog has said “AI-related issues arise” in its current court case against IAG over its allegedly misleading use of “an algorithm to determine a customer’s likelihood of agreeing to renew” their policy so that it could charge those more likely to renew higher premiums and vice versa.

QBE Australia made references to “the progressive roll-out of our underwriting AI Co-pilots” at its AGM last month but declined requests to elaborate on them.

Suncorp’s EGM of AI transformation Priyanka Paranagama told Information Age that “geospatial images of Australian homes combined with sophisticated AI” has been used “to simplify the quote process for customers, by shaving 50 percent of property questions needed to purchase home insurance.”

Non-traditional data feeds

Paranagama said that Suncorp’s AI solution for automating quotes was achieved “using aerial imagery to pre-populate information, like roof material and property features such as having a swimming pool or solar panels.”

Insurers are harnessing other non-traditional data sources to make pricing more automated and personalised, including IAG-owned NRMA, which lets car insurance members opt-in to a feature that collects their “location data”, and indicators of how safely they drive like “acceleration, braking, turning, speeding and mobile device use.”

Life insurer MetLife Australia's project manager Kate Kerr has used the example of using “productivity” and “absenteeism” to make predictions about health and referred to “coming together with the health industry, having it assess the way you exercise, where you go, how much you travel, how much you drive.

“This will all feed into a model where we can price risk according to the individual."

Woolworths’ Everyday Rewards card users who also purchase the supermarket’s car insurance, underwritten by Hollard Insurance, can get “a reduced premium” if they are predicted to be safe drivers based on “products purchased, price paid, time and location of shops”.

Claims processing and assessing with AI

MLC Life Insurance uses an assistive “dynamic triage model”, trained on customer data like “file notes” and “phone calls”, that “can figuratively tap the claim handler on the shoulder when a claim’s status changes” or if they have “missed a critical piece of information.”

In 2017, AIA Group reported using IBM’s Watson AI platform to achieve a "40 per cent reduction in turnaround times and over 99 percent accuracy on claims eligibility processing in Australia.”

Two years later, AIA Australia trialled another machine learning system that uses “comparative claims data to help speed up and streamline the claims process, comparing newly submitted claims against a bank of 60,000 historic claims.”

However, the company would not comment on whether either system was currently in use.