The Mills Review and the AI Personalisation Debate
A new review commissioned by the Financial Conduct Authority (FCA) has reopened a long-running debate: could increasingly sophisticated AI-driven pricing weaken the traditional risk pooling that underpins insurance? The Mills Review, published this month, flags concerns that as carriers become better at pricing individual risk using granular data, the cross-subsidisation between policyholder groups may erode, threatening broad access to cover.
Adam Pichon, senior vice president of global analytics at LexisNexis Risk Solutions, agrees that AI will make underwriting more sophisticated, but argues the technology is unlikely to dismantle the fundamentals of insurance. His assessment, shared with Insurance Business, highlights a more nuanced reality. While the review raises legitimate questions about fairness and access, Pichon believes practical limits—especially around data and the inherent randomness of claims—will prevent the kind of infinite personalisation that would truly break risk pooling.
Why Practitioners Say Risk Pooling Is Here to Stay
The Data Constraint
Pichon points to a hard practical limit: predictive models are only as good as the data they are trained on. In lines with limited claims history or sparse customer data, AI cannot segment risk beyond what the data supports. This is particularly relevant for commercial and specialist lines, where underwriting is still manual. Even in highly automated markets like motor, AI mainly improves back-end data processing and model fitting rather than enabling completely new rating factors.
The Randomness Factor
A second, deeper reason for residual pooling is that a significant portion of claims are fundamentally random. Insurers' models isolate the non-random part of risk, identifying groups with higher relative risk. But no model can predict individual claims with certainty. This irreducible randomness forces some level of aggregation, meaning cross-subsidisation will survive regardless of how advanced AI becomes.
Regulatory Watchpoints
The Mills Review places the tension between pricing accuracy and broad market access squarely on the regulatory agenda. It calls on insurers and supervisors to monitor whether hyper-personalisation could lead to unaffordable premiums for certain groups. However, Pichon's technical arguments may reassure regulators that the industry's own data and modelling constraints already act as a natural brake. The real challenge will be ensuring that AI-driven segmentation enhances fairness—by charging risk-appropriate prices and rewarding mitigation—without profiling in ways that are ethically or legally unacceptable.
Implications for the Market
Rather than revolutionising the pricing model, AI is likely to help carriers apply existing underwriting principles more effectively. Pichon notes that better-performing models let insurers identify genuinely higher-risk policies and price them accordingly, avoiding blunt rate increases across the board. This can improve profitability while creating a more resilient marketplace. For incumbents, the takeaway is that AI is a tool for better risk selection and operational efficiency, not a force that will atomise the entire pool.
What This Means for Insurers and Regulators
- For insurers: Invest in AI to speed up underwriting, improve the quality of manual lines, and refine risk segmentation—but don't expect to price every individual perfectly. Use better models to move away from broad rate hikes and toward risk-appropriate premiums, which can enhance fairness and retention.
- For regulators: Calibrate oversight to the practical limits of personalisation; data scarcity and claims randomness already put a floor under risk aggregation. Focus on ensuring that AI-driven pricing does not unfairly penalise vulnerable groups and that consumers who actively mitigate risk see real benefit in their premiums.
Risk & Opportunity Assessment
| Commercial Risk | Low | AI is an evolutionary improvement to existing underwriting, not a paradigm shift that threatens the viability of current business models; the industry can adopt it incrementally. |
| Competitive Risk | Medium | Carriers that effectively deploy AI for superior risk selection and operational speed could gain market share, particularly in lines where manual processes still dominate. |
| Regulatory Risk | Medium | The Mills Review signals regulatory attention; future rules on fairness and pricing transparency could impose new constraints, though the scope and timeline are uncertain. |
| Reputation Risk | Medium | If AI-driven personalisation is perceived as excluding higher-risk customers from affordable cover, public trust could suffer, though data limitations may blunt this effect. |
| Technology Disruption | Medium | AI enhances existing predictive models but does not fundamentally replace the pooled risk mechanism; the disruption is in process efficiency rather than the core insurance structure. |
| Commercial Opportunity | High | Better segmentation allows profitable underwriting of risks that were previously priced bluntly, opening up new growth avenues and reducing adverse selection. |
Comments 0