Insights

Life expectancy has become a systems problem

Life expectancy volatility is back in the headlines, and not for the reason most insurers want.

In the U.S., life expectancy at birth rose to 78.4 years in 2023 (up 0.9 from 2022) and then to 79.0 years in 2024 (up 0.6 from 2023). (CDC)

That sounds like stability returning.

But underwriting leaders and actuaries know the real story is not “people are living longer.” The story is that life expectancy has become a moving target, and the swings have been historically sharp. The Centers for Disease Control and Prevention (CDC) notes the U.S. experienced the two largest single-year drops in life expectancy in 2020 and 2021 since 1943.

That is not a trend line you can comfortably price with tables updated annually.

DIVIDER

Your biggest underwriting risk is not mortality. It is model decay.

When life expectancy trends change faster than your underwriting models and workflows, you get a hidden tax: mispricing, adverse selection, capital friction, and regulatory exposure.

That is why AI underwriting in insurance is moving from “innovation” to “control system.” Life expectancy is fragmenting across cohorts, geographies, and behaviors, creating risk identification patterns that static underwriting models were never designed to manage.

Globally, the volatility is visible. The Organization for Economic Co-operation and Development (OECD) reports that COVID‑19 drove around 6 million excess deaths across OECD countries in 2020–2022. As a result, average life expectancy fell by 0.7 years between 2019 and 2021, reversing years of progress across advanced economies.

This was not evenly distributed by country, age, or population group.

Even after the acute shock, the long tail remains uncertain. The Society of Actuaries (SOA) notes that mortality rates have declined significantly, but emerging data through still suggests a small amount of excess mortality for the 65+ population.

That combination, rebound plus residual uncertainty, creates a pricing and underwriting reality that looks like this:

• Different cohorts behave differently. Age bands are not enough.
• Risk drivers shift mid-cycle. What mattered two years ago may not matter now.
• Experience emerges faster than governance. Your business wants faster decisions, your risk team wants slower deployment.

DIVIDER

Why static underwriting models are failing

Traditional underwriting was built for a world where risk changed slowly and model updates were episodic. That world is gone.

Static models break in predictable ways:

They drift without warning

Even strong models degrade as population health, behavior, and treatment patterns shift. The impact often appears only after loss experience emerges, by which point the risk is already on the books.

They penalize speed

As uncertainty rises, teams add friction. More manual review. More checks. Longer cycle times. The result is slower growth and weaker distribution economics.

They increase compliance risk

As models grow more complex and opaque, decisions become harder to explain, audit, and defend.

This is where most “AI in underwriting” conversations stop, at automation. That is table stakes.

DIVIDER

The new operating model: Continuous underwriting

Think of underwriting like a flight system. Static underwriting flies with last year’s weather map. It works, until it doesn’t.

Continuous underwriting is the alternative. It uses AI and near-real-time data to continuously update risk assessment and decisioning as conditions change.

At a practical level, it requires three capabilities.

1. Dynamic risk scoring insurance can trust

Underwriting scores cannot be one-time outputs. They must be monitored, tested, and recalibrated through:

• Model performance monitoring
• Data and outcome drift detection
• Champion-challenger testing
• Defined thresholds for escalation to human review

This is how AI-driven underwriting models move from “smart calculators” to managed risk assets.

2. Decisioning that blends prediction and policy

Underwriting is never purely predictive. It combines analytics with rules, appetite, and compliance.

Modern stacks integrate:

• Predictive underwriting AI for risk estimation
• Rules engines for eligibility and thresholds
• Human underwriter review for accountability and edge cases

3. A feedback loop into pricing, not just underwriting

Underwriting can only move as fast as pricing allows. CIOs and CTOs should treat underwriting signals as upstream inputs to pricing governance, product design, and reinsurance strategy.

This is where longevity risk extends beyond annuities. When mortality and longevity patterns shift unpredictably, every long-duration promise is exposed.

DIVIDER

AI is advancing faster than underwriting controls

The market signal is clear, not from adoption statistics, but from operating behavior. Insurers are actively exploring generative AI across service, operations, and decision support, while underwriting remains more cautious.

That caution is warranted.

Underwriting is not a front-office productivity use case. It directly affects pricing integrity, fairness, regulatory defensibility, and long-term capital outcomes. Decisions made here must be explainable, repeatable, and auditable.

This is where many AI initiatives stall. Not because the models do not work, but because the controls around them are incomplete. Governance, monitoring, and exit strategies are often treated as follow-on activities. In underwriting, they are the enabling layer.

The real risk is not deploying AI too slowly.
The real risk is deploying AI faster than your underwriting controls can support.

That is why the insurers making progress are not leading with algorithms. They are leading with architecture, governance, and operating discipline.

DIVIDER

A CIO and CTO playbook for underwriting modernization

Modernizing underwriting without destabilizing the core requires more than point fixes. It requires sequencing — knowing what to modernize first, what to standardize, and what to connect across the enterprise.

The following five moves reflect the underwriting portion of a broader operating model outlined in UST’s Insurance AI Flight Plan — a practical guide for CIOs and CTOs on moving from pilots to enterprise performance across underwriting, claims, and operations.

Move 1: Instrument the underwriting data pipeline

Every modernization effort depends on this foundation. Without consistent, decision-ready data, AI cannot scale safely. This is where most underwriting AI initiatives quietly stall.

That means:

• Shared identifiers across applicant, policy, and claims outcomes
• Standardized intake for electronic health records and third-party data
• Feature stores with clear lineage and reuse

Move 2: Treat model monitoring as a product

If you cannot observe model health, you cannot control risk. This turns underwriting models into governed enterprise assets, not black boxes.

Leading insurers operationalize monitoring by defining:

• Accuracy and stability metrics tied to business outcomes
• Fairness and bias checks
• Audit trails and decision explainability
• Clear thresholds for escalation to human-in-the-loop underwriting

Move 3: Deploy AI with bounded autonomy

Avoid big-bang rollouts. Progress comes from targeted deployments where outcomes are measurable, risk appetite is clear, and human judgment remains available.

This approach delivers proof without putting the portfolio at risk.

Move 4: Connect underwriting intelligence to pricing and governance

Underwriting modernization fails if pricing cannot respond. Risk signals must flow into pricing, product, actuarial, and risk teams so decisions remain aligned across the enterprise. This is how AI moves from tactical automation to strategic decisioning.

DIVIDER

Conclusion

Life expectancy volatility is outpacing underwriting systems built for stability. When models lag reality, risk goes unseen, decisions become harder to defend, and exposure accumulates before anyone can act.

Winning insurers treat underwriting as a continuously governed decision system. UST’s Insurance AI Flight Plan shows CIOs and CTOs how to build it, sequencing data, governance, and workflows so underwriting keeps pace with a world that won’t slow down.

Explore the fight plan here.