The FCA-commissioned Mills Review has reignited debate over whether AI-driven pricing could reshape one of insurance's oldest principles: risk pooling.
Published this month, the review highlights stakeholder concerns that increasingly granular data and AI-driven pricing could weaken the cross-subsidisation that has traditionally helped spread risk across groups of policyholders. As insurers become better at pricing individual risk, it notes, the balance between pricing accuracy and broad access to insurance could come under increasing pressure.
Adam Pichon, senior vice president of global analytics at LexisNexis Risk Solutions, agrees AI will make underwriting more sophisticated, but argues the technology is unlikely to dismantle the fundamentals of insurance.
"It is more of an evolution," he said.
Pichon said AI's impact differs depending on the type of insurance being written. For commercial and specialist lines where underwriting and pricing still involve significant manual work, AI helps underwriters process information more efficiently.
"For types of insurance where underwriting and pricing involve a manual process (for example, some property lines and commercial insurance lines), AI improves access to and summarisation of information,” he said. “This will ultimately yield faster underwriting and higher quality underwriting."
In markets where underwriting is already highly automated, such as motor insurance, the benefits are different. AI enables faster, higher-quality back-end data work, quicker model fitting and improved predictive models, helping insurers improve pricing accuracy rather than fundamentally changing how they assess risk.
Pichon said the same neural networks underpinning large language models can also improve deterministic models, making them more accurate without necessarily leading to greater personalisation.
The Mills Review raises concerns that increasingly personalised pricing could reduce the level of cross-subsidisation between customer groups. Pichon believes there are practical reasons why insurance is unlikely to become infinitely personalised; the first is data.
"Predictive models are constrained by data. If there is limited data to train a model, there are limitations in how granular that model's predictions can be,” he said.
Pichon said this creates a practical limit on how personalised insurance pricing can become because there will always be limits to how finely risks can be segmented.
The second constraint is that some insurance claims are inherently unpredictable.
"A percentage of insurance claims are, fundamentally, random events. Insurers' models are designed to identify increased relative risk – essentially, to isolate the 'non-random' part of risk," he said.
"But a large proportion will always be random, which means some level of risk aggregation will always occur. Put another way, no insurer model will ever predict with 100% accuracy who will and won't have a claim because some portion of risk is simply random."
Instead of creating an entirely new pricing model, Pichon said AI is more likely to help insurers apply existing underwriting principles more effectively.
He said better-performing models allow insurers to identify which policies genuinely present higher levels of risk and charge risk-appropriate prices, rather than relying on broad rate increases when claims costs rise.
This more accurate segmentation enables insurers to operate profitably while creating a more efficient and resilient insurance marketplace. He also argued it can improve fairness, accessibility and affordability for consumers who actively mitigate their risks.
The Mills Review highlights increasingly personalised pricing as one of the questions regulators and insurers will need to navigate as AI adoption accelerates. Pichon's assessment suggests the debate may be less about whether AI can reshape traditional risk pooling than about the practical limits of personalisation.
While underwriting models are becoming more sophisticated, finite data and the random nature of claims mean some degree of risk aggregation is likely to remain fundamental to insurance.
