How AI Is reshaping fraud detection, insurance analytics and data governance

Towards the end of 2025, the Risk Management Practice Committee (RMPC) brought together experienced analytics specialists from AUSTRAC, Suncorp and the University of Melbourne for a panel discussion at an Actuaries Institute Insights Session titled Actuaries, AI, and Analytics: Strengthening Financial Crime Defences.

Speakers Ignatius McBride, Carly Bergmann and Dr Han Li, moderated by Clement Peng, examined how artificial intelligence, alternative data and improved governance are transforming fraud detection and insurance operations in Australia, highlighting both the opportunities and the risks.

From detection to disruption: how is AI changing fraud prevention?

A key theme was the opportunity for institutions to move beyond simply detecting fraud to actively disrupting and deterring it. AI is now improving the quality and usefulness of vast, complex datasets. Extractive AI tools enable unstructured documents to be transformed into practical insights, speeding up the identification of suspicious networks and persons of interest.

These advanced capabilities are making public–private partnerships — between financial institutions and government agencies — increasingly important avenues for jointly analysing data, generating leads and responding quickly to serious crimes like child exploitation and organised fraud. The panel stressed that this shift from reactive detection to proactive disruption represents one of the more significant changes in how financial crime is being addressed across the sector.

What is behavioural biometrics and why does it matter for fraud detection?

The rapid growth of behavioural biometrics was a standout topic from the session. Rather than analysing what users type, emerging tools analyse how they type — patterns in keystroke cadence, navigation behaviour and hesitation around key personal details. These signals, captured at high frequency during online interactions, offer a powerful complement to traditional fraud detection methods and are proving difficult for bad actors to replicate consistently.

The panel noted, however, that AI tools are as readily available to malicious actors as they are to regulators and law enforcement. The same methods used to detect fraud, including behavioural biometrics, can be used to evade detection. This creates a dual dynamic: traditional fraud risk decreases as detection improves, while entirely new categories of risk emerge that require ongoing vigilance and adaptation. The arms race between fraud prevention and fraud perpetration is, in this sense, accelerating rather than resolving.

How are insurers using alternative data to improve decision-making?

From an insurance perspective, the use of alternative data is expanding well beyond what has traditionally been available to underwriters and analysts. Beyond claims and policy data, insurers are increasingly leveraging publicly available information, such as visa statistics and international arrivals data, to forecast sales volumes in private health insurance. Customer engagement data from wellness programs and mobile apps is providing new insight into policyholder behaviour, helping insurers improve retention and better understand lapse patterns.

The panel noted that many of these insights emerge not from sophisticated new data collection, but from recognising value in datasets that already exist and have been largely overlooked. The practical implication is significant. Organisations that invest in data literacy and cross-functional collaboration are likely to surface meaningful signals well ahead of those that wait for purpose-built solutions.

What role can AI play in claims management?

Claims management was highlighted as one of the areas where AI has the greatest practical potential in insurance. The traditional claims process can be frustrating for customers — involving repeated requests for information, multiple exchanges and decision delays on more complex cases — and costly for insurers, who must manually review large volumes of documents, often including irrelevant material.

AI can support claims managers by automating document sorting, information extraction and summarisation, helping to speed up decisions and improve the customer experience. This also frees claims managers to focus attention on complex cases and vulnerable customers who need greater care and human judgement.

Nonetheless, strong quality assurance and oversight remain essential. Incorrect claim decisions, whether wrongly declining a genuine claim or accepting an invalid one, can cause direct harm to claimants and, if left unaddressed, affect the affordability of insurance more broadly through higher premiums. The efficiency gains AI offers are real, but they do not reduce the responsibility that sits with the people making decisions.

How should organisations manage model risk and AI governance?

A recurring message from the panel was that AI should be viewed no differently from any other actuarial model as it introduces model risk and must be governed accordingly. A key pitfall raised was the inability to explain AI outputs, which is itself a material risk. AI can also produce results with high apparent confidence when the underlying prompt quality is poor, making validation discipline critical.

As with traditional actuarial models, strong governance frameworks and independent validation remain essential. The discussion also highlighted growing concern about data privacy, the ethical use of third-party data and risks such as model drift — where a model that performs well at deployment gradually degrades as real-world conditions shift. The panel's view was clear. Industry guidelines, privacy impact assessments, transparency statements and continuous monitoring are not optional extras but foundational requirements for safe AI adoption.

What does AI mean for the future of actuaries?

The speakers were direct in dismissing the idea that AI would displace actuaries. AI still requires people who are skilled and knowledgeable in the domain in which it is being applied and that is precisely where actuaries are well placed. The growing emphasis on model oversight, explainability and governance points to a shift in how actuarial work is performed, with less focus on manual computation and more on holding models accountable and ensuring their outputs can be interpreted and communicated clearly.

Given that interpretation and explanation are core actuarial competencies, the profession is well positioned to take a leading role in responsible AI adoption across financial services. The consensus from the panel was consistent: AI should augment professional judgement, not replace it. It can increase confidence in decision-making and improve efficiency, but accountability must always remain with the actuary.

What's next

Conversations like this one happen regularly across the Actuaries Institute community. Whether you're working in risk, data science, insurance or beyond, there's a session for you — explore what's coming up through the events calendar.