An insurer can tell a customer exactly what their premium is, but explaining why it landed at that particular number can be considerably harder.
That gap is becoming more consequential as underwriting draws on wider sets of personal, behavioural and third-party data, while artificial intelligence gives insurers greater capacity to combine and analyse those inputs.
It is a tension examined by Fei Huang, associate professor at the UNSW School of Risk and Actuarial Studies, who has raised questions about how much information can now sit behind insurance pricing and whether customers can see enough of the process to understand the result.
Depending on the cover, insurers may use information including age, health, driving history, property characteristics and claims history. Huang noted that relationship data such as payment frequency and policy duration can add another layer, alongside publicly available information.
Beyond that sits a much larger data industry. The ACCC has documented firms offering products built from demographic, purchasing, location and online behaviour, including risk and fraud-management products used for insurance applications. It has also warned that consumers often have limited visibility over how information is collected and combined, and that de-identified information can potentially become identifiable when linked with other datasets.
That does not mean every insurer is using all of this information. But the pool of data capable of informing risk assessment is expanding as technology makes it easier to process.
AI-assisted underwriting can bring together fragmented information, identify patterns across portfolios and support faster decisions. That can improve segmentation, but it also means two insurers assessing the same customer may reach different conclusions based not only on appetite, but on the data they hold and how their models interpret it.
There are already questions over how well those decisions can be explained. A General Insurance Code Governance Committee review of online motor applications across 13 insurers and 58 brands found some insurers could not demonstrate how certain questions were relevant to their decisions. Customers declined cover were also often given vague explanations.
The issue becomes more sensitive when apparently neutral information correlates with attributes protected under discrimination law.
Huang highlighted the risk of proxy or indirect discrimination, where one variable effectively stands in for another characteristic. Removing a protected attribute from a dataset does not necessarily remove the problem if other information produces similar segmentation.
Guidance from the Australian Human Rights Commission and Actuaries Institute makes a similar point, warning that insurers relying on anti-discrimination exemptions must be able to justify the data underpinning their decisions. It cautions against relying on information that is “out-of-date, qualified, incomplete, discredited, based on an insufficient sample size, or not directly applicable to the particular situation.”
The issue is now colliding with a separate push for greater pricing transparency.
ASIC’s August review of motor insurance premium transparency covered five insurers and eight brands representing about 72% of the market. None of the eight brands explained in their quote and renewal documents “the key factors that affected the calculation of the premium, or why the premium had changed from the previous year.”
Motor premiums rose 8% in the 12 months to July 2025 after increasing more than 42% between 2019 and 2024. More strikingly, 31% of consumers who contacted their insurer before renewing secured a lower premium without changing their cover, while 68% did not contact their insurer at all.
That gives transparency a practical dimension: a renewal figure may be difficult to unpack, but questioning it can still change the outcome.
Financial Services Minister Daniel Mulino has also called for clearer explanations of material premium increases and greater consistency in how insurers communicate pricing.
Regulatory attention is extending to the technology behind those decisions. APRA called for a “step-change” in AI risk management in April after finding governance, assurance and operational resilience were not keeping pace with AI adoption among large banks, insurers and superannuation trustees.
The boundaries of underwriting data are changing too. From October 8, new legislation will restrict life insurers from using protected genetic information in underwriting. From December 10, insurers covered by the Privacy Act will also face new transparency requirements where personal information is used in significant automated decision-making, though the rules do not require disclosure of the mechanics of a pricing algorithm.
The result is a more complicated underwriting environment, where access to data is expanding at the same time as the acceptable use of that data is being more closely defined.
