{{item.videoDuration}}
{{item.title}}
{{item.text}}
{{item.videoDuration}}
{{item.text}}
Key takeaways:
- A thoughtful approach to AI involves making disciplined investments that promote measurable improvements in underwriting, pricing, and claims decisions.
- AI models don’t create value by themselves. The operating system around them—the measurement, compliance, and control system—makes AI dependable, auditable, and economically measurable.
- Evaluating AI through unit economics, attribution, and governance can move carriers from experimentation to an engine for margin improvement, operating leverage, and stronger returns at scale.
Many P&C insurers’ economic focus on AI starts in the wrong place—token cost and cloud usage. But they're usually a rounding error compared to the financial impacts of claims settlement and pricing adequacy.
You instead should concentrate on how AI can produce measurable, durable improvements to the business.
How AI can benefit underwriting, pricing, and claims
P&C carriers’ largest near-term opportunities to benefit from AI lie in pricing, underwriting, and claims—functions that directly shape loss, expense, and ultimately combined ratio. We recommend assessing AI not by cost per prompt but by net value in the following business transactions.
- Underwriting: submission, quote, bound policy, renewal, referral.
- Pricing: rating indication, rate adequacy review, rate change, renewal price action right.
- Claims: type of claim (e.g., litigated), feature, closed claim.
Underwriting selection and submission quality (loss and expense ratio improvement)
Underwriting is where exposure selection, pricing inputs, capacity deployment, and risk appetite meet. Small improvements in selection and documentation can compound across the portfolio. AI can help you by:
- Extracting structured exposure characteristics from unstructured submissions (applications, schedules, loss runs).
- Improving appetite alignment, referral quality, authority checks, and guideline consistency.
- Reducing rework, missing-information cycles, and quote-to-bind friction.
- Standardizing documentation of underwriting judgment and exceptions.
Pricing adequacy and rate execution (loss ratio improvement)
Pricing value comes from a fast, disciplined pricing feedback loop with effective indication, segmentation, and rate action prioritization, as well as quick course correction. AI can raise pricing quality by improving the consistency and completeness of risk signals by:
- Detecting emerging rate inadequacy by segment, geography, peril, coverage, channel, or risk characteristic before it is fully visible in aggregate loss ratios.
- Improving segmentation by identifying interactions among exposure, claims, and external data that may not be captured in current rating structures.
- Prioritizing rate actions by estimating where pricing changes are most likely to improve loss ratio, retention, new business conversion, and portfolio mix.
- Reducing misclassification and “defaults” in rating variables.
- Standardizing guideline interpretation and documentation of pricing decisions.
- Enabling faster iteration from indication to filed/runnable implementations (with controls and filing documentation meeting regulatory compliance requirements).
Claims leakage reduction (loss ratio + allocated loss adjustment expenses (ALAE))
Leakage improvements can be measured per claim and validated through controlled rollout. AI can help reduce avoidable loss by:
- Verifying coverage and limits, and structuring extraction from policies and claim documents.
- Identifying billing anomalies, overpayment, missed recoveries, duplicate services, inflated estimates, or inconsistent narratives.
- Surfacing subrogation opportunities earlier and prioritizing recovery actions.
- Providing triage and next-best-action guidance to reduce missed steps.
Cycle time compression (severity, ALAE, litigation propensity)
Claims performance is sensitive to speed and handling quality. AI can help reduce cycle times—thereby lowering ALAE and (depending on business line and jurisdiction) severity creep. AI-enabled benefits include:
- Better summaries and issue identification that decrease rework and escalations by promptly surfacing key facts, coverage questions, reserve issues, missed documentation, and settlement considerations.
- Faster decisions that reduce both late-stage surprises and prolonged open claims by accelerating triage, referral, investigation, and next-best-action decisions.
- Improved customer communication which reduces the friction that can drive litigation, regulatory scrutiny, or bad-faith exposure by supporting clearer explanations, timely updates, complaint-risk identification, and consistent documentation.
Expense ratio improvement (capacity gains)
Time saved is not ROI unless it lowers overtime, reduces vendor spend, shortens cycle time, increases quote/claim volume with the same or fewer staff, and measurably improves quality. AI can create value by increasing throughput per adjuster/underwriter in a number of areas.
- Intake, document understanding, first notice of loss (FNOL) triage.
- Drafting correspondence, agenda management, and structured note generation.
- Reducing “time to competence” for new staff via embedded expertise.
Small, measurable margin improvements in loss and expense ratio, claims leakage, cycle time, and capacity can create meaningful economic impact when deployed at scale and with disciplined use of the right units of measurement. And once you identify these units of measurement, AI use can become legible in business rather than tech terms: margin, growth, and risk.
Small, measurable margin improvements can create meaningful economic impact at scale.
Actuarial-grade attribution means credible—not anecdotal—measurement
C‑suite teams are right to be skeptical of ROI claims based on demos and promises of time saved. You need to be able to confirm value with model designs that control for trend, mix, and randomness.
Effective measurement patterns
- A/B testing: Handlers or underwriters are randomly assigned AI assistance versus control (where operationally feasible).
- Stepped-wedge rollout: Regions/teams adopt over time, enabling before/after comparisons while controlling for calendar effects.
- Matched cohorts/propensity matching: Compare similar claims/risks to isolate AI impact when randomization isn’t possible.
- Difference-in-differences: Separate AI impact from macro drivers (inflation, legal environment, catastrophe activity, portfolio mix shifts).
Measurement needs to include leading and lagging indicators.
- Leading: Touch rate, referral rate, cycle time, subrogation identification rate, exception rates, override rates.
- Lagging: Paid severity, indemnity/ALAE per claim, ultimate loss emergence, reopen rate retention, and rate adequacy metrics.
This is the discipline that converts “AI optimism” into credible financial guidance.
Calculations for evaluating AI ROI |
To evaluate AI like any other investment, express benefits as incremental cash flows. For period t: Incremental cash flow(t) Premium uplift(t) + Loss reduction(t) + ALAE reduction(t) + Operating expense reduction(t) + Capital cost reduction(t) – AI costs(t) Then compute:
This is the bridge from model usage to company value. But the model works only if you can credibly estimate cash flows, and the most effective way to do that is through unit economics. Defensible value measurement for business unit economics Pick a unit and compute: Net value per unit = (Expected margin improvement per unit) – (AI variable cost per unit) – (incremental human review cost) Then scale: Annual net value = volume × net value per unit – Fixed program costs (amortized) This approach does two important things for leadership teams.
|
Reference: AI costs |
Program (fixed costs) |
|
Run (variable costs) |
|
Effective governance and risk management drive long-term value
The fastest way to destroy ROI is to treat AI as a chat layer rather than an engineered control system. And that system should include a model inventory, approved-use register, data lineage, validation standards, monitoring, change control, incident management, audit trails, and third-party oversight. Two insurers can spend the same on AI and get opposite outcomes depending on how they embed those controls into the operating model.
For executive teams, the must-haves are straightforward.
- Observable: Instrument AI workflows so you can see evidence, outputs, decisions, overrides, exceptions, latency, and cost.
- Measurable: Evaluate AI against clear business metrics, including accuracy, workflow outcomes, financial impact, and customer/compliance guardrails.
- Repeatable: Use structured test sets, regression testing, and version control so you can consistently compare prompt, model, data, and workflow changes.
- Stable: Monitor production performance for drift, errors, regressions, complaint signals, and outcome deterioration.
- Compliant and controlled: Maintain auditable decision rationales, human oversight, bias/adverse-impact monitoring, privacy and security controls, regulatory documentation, and third-party oversight.
- Integrated: Connect AI outputs into your underwriting, claims, and pricing systems to reduce re-keying, ambiguity, and off-system decisioning.
The model doesn’t create value by itself. The operating system around it—the measurement, compliance, and control system—makes AI dependable, auditable, and economically measurable. The mandate is simple: Measure twice, automate once.
The fastest way to destroy ROI is to treat AI as a chat layer rather than an engineered control system.
Sequencing your AI strategy
An effective AI strategy is not a single roll of the dice, but a disciplined, long-term portfolio of initiatives. Sequencing is critically important. Early wins generate funding, build confidence, and reduce execution risk for subsequent, higher-value opportunities. Here’s an approach you can use to promote stability, capital efficiency, and strategic agility.
Two insurers can spend the same on AI and get opposite outcomes depending on how they embed controls into their operating model.
Conclusion: AI as a margin lever
For P&C executives, a thoughtful approach to AI involves making disciplined investments to achieve measurable business outcomes. Specifically, your AI tools need to improve underwriting, pricing, and claims decisions—and therefore financial performance.
When you evaluate AI through unit economics, attribution, and governance, you can move from experimentation to enterprise value. That’s when AI becomes more than a technology initiative and turns into an engine for margin improvement, operating leverage, and stronger returns at scale.
AI Risk Modeling
{{filterContent.facetedTitle}}
{{contentList.loadingText}}
{{contentList.loadingText}}
Follow us
