Two carriers writing the same segment at similar rates are not competing on equal terms if one can see risk variation within that segment and the other cannot. The carrier with the better model prices below-average risks competitively. The carrier without it wins the above-average risks by default — not because it sought them, but because it could not distinguish them.
That dynamic plays out across thousands of renewals without any single decision appearing obviously wrong. The loss ratio deteriorates gradually. By the time it is visible the book has already drifted, and rate action alone does not fully correct a portfolio whose composition has shifted because the underlying selection mechanism has not changed.
Pricing precision is one lever on book quality alongside underwriting rules, appetite management, and distribution relationships. But it is the lever that operates at scale on every risk in the book at every renewal.
Where the gap opens
Actuarial tables compress heterogeneous risk populations into segments defined by observable, defensible variables. That compression is necessary for regulatory tractability and actuarial certification. It also means the table prices two risks identically when their actual loss propensity is materially different.
Two drivers in the same age band and territory may behave very differently behind the wheel. Two properties in the same postcode and construction type may have very different weather exposure depending on roof condition and drainage. The table cannot see those differences. Where competitors can, the carrier that cannot loses the better risk and wins the worse one on every such pairing.
A 1-point combined ratio improvement on a $1 billion book is worth $10 million annually. That improvement does not require writing more premium. It comes from pricing the existing volume more accurately — retaining the risks whose modelled loss expectation is below the rate they will pay, and releasing the risks whose modelled loss expectation exceeds it.
The data that supports more precise rating
The signals that distinguish individual risk within a segment are available at the point of quote across most lines of business.
In personal auto, telematics data captures actual driving behaviour — miles driven, time of day, braking patterns, route types — that predicts individual claim frequency and severity more accurately than the demographic and vehicle variables that define the segment.
In personal property, aerial and satellite imagery provides condition-level signals on individual properties. Roof condition scored from imagery predicts weather-related losses more accurately than roof age from application data, because the model observes actual condition rather than the recorded installation date.
In commercial lines, third-party data enrichment adds operational and financial signals that are not on the application but that are available from company data providers. An account with rapid growth and high credit utilisation presents a different risk profile from a stable account in the same underwriting category, even when application data treats them equivalently.
The competitive consequence is straightforward. Carriers that incorporate these signals into their rating price the book more accurately. Carriers that do not price against segment averages and accept the adverse selection that follows.
The regulatory constraint
Rate filings require that factors are actuarially justified, explainable to regulators, and do not produce disparate impact on protected characteristics. That obligation applies to granular models as directly as it does to traditional tables. A variable that is highly predictive but correlates with a protected class will face scrutiny that can result in filing rejection or required remediation.
This does not prevent more precise rating. It means the model design, variable selection, and actuarial justification need to account for regulatory requirements from the start. Carriers that have treated explainability and proxy testing as afterthoughts have had programmes suspended. Those doing it well integrate compliance review into the model development process.
The technology dimension
At most large P&C carriers, the policy administration system holds the rating data, prior claims history, and customer profile that a granular pricing model needs at the point of quote. For carriers running their core policy and claims infrastructure on IBM Z, deploying the rating model via IBM Machine Learning for z/OS keeps the precision scoring within the rating transaction, with direct access to the full policy and loss history on the same platform. The model returns a granular risk score to the rating engine as part of quote generation without requiring data extraction or a round trip to an external scoring environment. For carriers on other infrastructure the same principle applies: the model needs to run where the quote is generated, with access to the data that supports it.
What success looks like
The relevant metrics are price adequacy by risk segment, loss ratio on the precision-rated cohort against the standard-rated cohort, and the within-segment distribution of bound business over time. That last measure captures portfolio drift directly. Loss ratios change as a consequence of composition shifts. Tracking composition gives earlier warning of whether the rating model is working before the loss experience confirms it.
