Insurers and risk functions weigh AI as analytical tool and fresh exposure

Artificial intelligence is being pitched as a remedy for the data and analysis bottlenecks in insurance and enterprise risk management, yet both accounts of that shift also record the exposures the technology creates. Viva reported that insurers and reinsurers face pressure to make more accurate, adaptive and forward-looking decisions as risks grow more complex, citing changing health behaviour, economic dynamics and climate impacts, and that conventional approaches built on historical data and manual analysis are becoming less adequate as data volumes and variety expand. Writing separately on corporate risk management, AI4Business framed the technology as dual-use: it offers more powerful tools for analysing, identifying and monitoring risk, but also introduces operational, IT, reputational and model-reliability vulnerabilities and can widen an organisation's attack surface.

The Indonesian account locates much of the problem in data plumbing rather than data scarcity. In many situations, it reported, information is available but held in separate systems, in formats that are not uniform and often only in aggregate form. AI can help standardise formats across sources, flag incomplete or inconsistent records and reduce reliance on manual input, while also detecting shifts in claims trends earlier, supporting risk classification and prioritising cases that need closer review. It could sharpen the comparison of actual portfolio experience against the assumptions used in risk management by combining historical data, portfolio characteristics and other variables, giving a more detailed picture of a changing risk profile so assumptions can be revisited sooner. The same account is explicit about the limits: performance depends on data quality, model accuracy and context, and models trained on incomplete, unrepresentative or biased data can produce faulty recommendations.

AI4Business drew its observations from a seminar on AI and generative AI for risk management and compliance organised by CETIF, attended by representatives of banking, insurance, academia and technology, with the article's author taking part as a speaker. The piece sets out three directions of evolution, of which two are visible in the supplied excerpt. The first is a move from qualitative to data-driven, forward-looking risk management: operational, cyber, supply-chain and reputational risks rarely come with ready-made historical series, and machine learning can help recognise patterns, correlations and weak signals, with applications running from cyber risk ratings to exposure quantification in which probability and frequency are expressed as intervals rather than certainties, allowing scenarios to be compared and investment prioritised. No model can guarantee prediction of an incident, the account notes, but a good one can shorten the interval between the first signal and a decision. The second direction concerns unstructured material — contracts, regulations, policies, reports, assessments and supplier documentation — which generative AI can bring into contract analysis, due diligence, the identification of regulatory gaps and the maintenance of control catalogues.

Neither supplied excerpt records a regulatory decision, an adopted standard or a timetable for one, and the AI4Business text breaks off mid-sentence before its third direction is developed. The open question both accounts leave is one of control: how organisations capture the analytical gains while governing the risks the tools themselves introduce.