Where the Money Is Going
Ask a P&C carrier where they are investing in AI and the answer, in most cases, is claims. Fraud detection. Reserve setting. Settlement optimisation. Injury severity scoring. Triage at first notice of loss. These are legitimate problems with genuine AI solutions, and the data that supports them — claims history, injury profiles, settlement outcomes — is among the richest in the industry.
The question worth asking is not whether claims AI creates value. It does. The question is whether the industry has thought carefully about where the most value sits — and whether the investment mix reflects that analysis or merely reflects where the data is easiest to work with.
Our view is the latter. P&C carriers have built their AI portfolios around the post-loss layer because claims data is rich, ROI calculations are tractable, and the business case is straightforward to make. The pre-loss layer — underwriting quality, pricing precision, and catastrophe accumulation management — receives a fraction of the investment, despite offering materially higher economic leverage. This is not a deliberate strategic choice in most carriers. It is the default that emerges when AI investment decisions are driven by data availability and near-term ROI visibility rather than by combined ratio impact.
The Compounding Argument
The distinction between pre-loss and post-loss AI is not merely a matter of timing. It is a matter of what the investment is actually doing.
Claims AI improves the management of losses that have already been written. A better fraud detection model recovers more of what was paid fraudulently. A better reserve-setting model reduces the adverse development on open claims. A better settlement model reduces leakage on the claims being closed. These are genuine improvements. They operate on losses that are already in the book.
Underwriting AI improves the quality of the risks that are written. A better risk selection model declines submissions that would have become adverse claims. A better pricing model charges premiums that accurately reflect expected loss on every risk in the portfolio. A better renewal model retains the profitable risks and exits the unprofitable ones. These improvements do not operate on losses already in the book — they change what enters the book in the first place.
The combined ratio arithmetic makes this distinction stark. A one-point combined ratio improvement on a $1 billion gross written premium book is worth $10 million in underwriting profit. That improvement, if achieved through underwriting AI, compounds over the life of every policy written more accurately — three, five, ten years of portfolio quality gains that continue to accumulate as the book renews. The same $10 million achieved through claims AI is recovered on the losses in the current year. Next year, the claims portfolio resets. The underwriting quality gain does not.
This is not an argument against claims AI. It is an argument that underwriting AI should command a higher investment priority than the current portfolio mix in most carriers reflects.
Where the Pre-Loss Leverage Is Greatest
Not all pre-loss AI investments are equal. Two domains stand out as carrying disproportionate leverage relative to current investment levels.
Pricing precision is the highest-leverage underwriting application in a competitive market — and the one most consistently underinvested relative to its economic return. The mechanism is competitive rather than absolute: a carrier that prices risk more precisely than its competitors does not simply achieve a better loss ratio on a fixed book. It wins the best risks that competitors are overpricing, avoids the worst risks that competitors are underpricing, and compounds this selection advantage across every renewal cycle. A carrier with better pricing models is not competing for the same risks at better margins. It is operating in a different risk pool — one that systematically improves as competitors continue to misprice.
The carriers that have invested in granular pricing models — using telematics, imagery, third-party data, and behavioural signals to price individual risks rather than actuarial bands — have not just improved their own combined ratios. They have changed the composition of the industry’s risk pool in their favour. The risks they have declined or overpriced flow to competitors with less precise models. This is a compounding competitive dynamic that claims AI cannot replicate.
Catastrophe accumulation is the pre-loss domain where the cost of inadequate AI is most acute — and most episodic, which is precisely why it is underinvested. Between catastrophe events, accumulation risk is invisible. Individual policy underwriting decisions do not appear, in isolation, to be building dangerous geographic or peril concentration. The concentration is only visible at the portfolio level, and it is only consequential when an event occurs. By that point, the underwriting decisions that created the exposure have long since been made.
Carriers that manage cat accumulation with static models or periodic manual review discover their true PML at the point of loss, not before it. The reinsurance that should have been purchased was not purchased — or was purchased at the wrong attachment point — because the exposure was not accurately known in advance. A single miscalibrated PML on a major event can leave a carrier under-reinsured by tens of millions of dollars. The AI investment that prevents that outcome is not visible in the annual combined ratio in the years when no major event occurs. It is catastrophically visible in the year when one does.
Why the Post-Loss Bias Persists
The concentration of AI investment in claims is not irrational — it is the predictable result of how investment decisions get made inside carriers.
Claims data is richer, more structured, and more historically complete than underwriting data. The outcome labels that train supervised models — fraud confirmed, reserve adequate, settlement within range — are available in claims in ways that are harder to construct for underwriting decisions, where the outcome of a selection or pricing decision may not be visible for years. This makes claims AI faster to build, easier to validate, and more straightforward to present to a business case committee.
Claims problems are also more operationally urgent. A backlog of open claims, a rising LAE ratio, a fraud ring generating sustained losses — these are visible, immediate, and demand a response. The opportunity cost of imprecise underwriting is diffuse and slow. It does not generate an incident report. It does not appear in weekly operational metrics. It accumulates quietly in the loss ratio over years, surfacing only when the book has deteriorated enough to force a correction.
The result is that AI investment decisions in most carriers are driven by problem urgency and data availability — both of which favour claims — rather than by economic leverage, which favours the pre-loss layer.
The Recommendation
Audit your AI investment portfolio against the loss timeline.
Categorise every current and planned AI investment as pre-loss — improving decisions that determine what enters the book — or post-loss — improving the management of losses already written. Calculate the proportion of investment falling into each category. In most carriers, this exercise will reveal a significant imbalance toward the post-loss layer.
The rebalancing that follows should not defund productive claims AI. It should redirect incremental investment — the next programme, the next model build, the next data infrastructure project — toward pricing precision, underwriting quality, and catastrophe accumulation management. The business case for this rebalancing does not require precise ROI calculations. It requires the recognition that a dollar of combined ratio improvement achieved before a loss enters the book is worth more, over the life of the book, than a dollar recovered after one does.
The carriers that build that recognition into their AI investment frameworks will compound their underwriting quality advantage over time. The carriers that continue to invest downstream will continue to manage the consequences of decisions that better pre-loss models would have made differently.
