Insights

The insurers scaling AI aren’t building better models; they’re rebuilding the system around them

Prashanth Krishnamurthy – Insurance client partner

Most insurers believe underwriting modernization means deploying better predictive models.

That belief is increasingly dangerous.

The real competitive divide in insurance is no longer who has the most advanced algorithm. It is who has designed underwriting systems that minimize underwriting blind spots. And today, blind spots are expanding faster than most executive teams realize.

Across the industry, leaders expect AI to improve underwriting productivity and technical results. Yet only a small fraction of insurers have brought AI initiatives to enterprise scale. BCG reports that just 7% of insurers have successfully scaled AI, while roughly two-thirds remain in pilot mode. The constraint is not data science capability. It is operating model design.

Today’s underwriting still relies heavily on structured data, episodic model updates, and manual intake normalization. Risk, however, has evolved. It is multimodal, behavior-driven, climate-sensitive, and continuously shifting. When underwriting operates episodically while risk evolves continuously, blind spots accumulate quietly inside the system.

They do not show up immediately in quarterly performance. They emerge gradually through higher referral rates, inconsistent decisioning, growing audit scrutiny, and subtle loss ratio drift.

This is not a prediction problem. It is a systems problem.

DIVIDER

Predictive vs precision underwriting

The next era is not about marginally improving model accuracy. It is about redesigning underwriting as a precision decision system.

This shift requires a structured precision underwriting framework capable of integrating AI-ready underwriting data, decision intelligence, governance controls, and continuous monitoring into a unified operating model.

Precision underwriting represents a shift from proxy-based segmentation toward context-rich, continuously refreshed risk assessment. Gartner’s 2025 AI Hype Cycle highlights AI-ready data, multimodal AI, and AI agents in insurance as fast-advancing priorities — all directly implicated in underwriting pipelines that blend text, imagery, and sensor signals.

Three structural forces are driving this change.

1. Underwriting has become multimodal

Risk evidence no longer lives neatly in structured tables. It appears in PDFs, emails, aerial imagery, inspection photos, medical narratives, telematics feeds, and IoT sensor data.

Modern underwriting systems increasingly rely on image and text analysis for underwriting, allowing insurers to interpret inspection photos, engineering reports, property imagery, and submission documents as part of the risk evaluation process.

Climate volatility is also reshaping how insurers evaluate exposure. Increasingly, underwriting teams rely on climate risk underwriting models that integrate geospatial intelligence, catastrophe simulations, and environmental data to anticipate loss exposure before events occur.

All of these capabilities are increasingly powered by multimodal AI for insurance underwriting, which allows insurers to process and interpret diverse forms of risk evidence simultaneously.

BCG reports that in commercial P&C, underwriting efficiency in complex lines can improve by up to 36%, and loss ratio optimization can improve by incorporating previously inaccessible unstructured information. That gain is not about “better models.” It is about richer evidence.

If your systems cannot ingest and govern multimodal underwriting data reliably, you are underwriting partial reality.

2. Workflow redesign drives more value than model tuning

Most AI programs stall not because algorithms fail, but because underwriting workflows remain unchanged.

Industry research shows broad experimentation with generative AI across the sector, yet only a small percentage of insurers have achieved scaled AI in insurance at the enterprise level. BCG reports that just 7% of insurers have brought AI to scale, while roughly two-thirds remain in pilot mode. Implementation is common. Operationalization is rare.

The bottleneck is not model performance. It is workflow redesign, governance, and enterprise integration. Implementation is not transformation.

3. Governance is architectural, not procedural

Regulatory scrutiny is accelerating.

The NAIC Model Bulletin on the Use of AI Systems by Insurers makes clear that AI-supported decisions must comply with unfair trade practice and discrimination standards, supported by a formal written governance program.

In Europe, the EU AI Act treats AI used for life and health underwriting risk assessment as high-risk, triggering more stringent governance requirements.

This changes the scaling equation. If governance is layered on after deployment, scale slows. If governance is embedded in the architecture, scale accelerates. The competitive advantage is not explainability as a report. It is explainability as a system control.

Frameworks like the NIST AI Risk Management Framework (AI RMF 1.0) are increasingly used to structure this governance foundation.

DIVIDER

How to eliminate underwriting blind spots are actually created

Most modernization efforts focus on scoring. That is downstream. Blind spots are created upstream.

They originate in inconsistent submission intake, unstructured evidence that is never fully digitized, disconnected third-party data, weak entity resolution, and insufficient feedback loops from claims and renewals. Once flawed representations enter a model, no amount of algorithmic sophistication can fully compensate.

The emerging enterprise underwriting AI architecture reflects this reality. It emphasizes controlled ingestion, feature governance, explainability artifacts (e.g., SHAP-based attribution), human-in-the-loop review, and continuous monitoring.

This is not theoretical.

It is the difference between managing a model and managing a decision system.

DIVIDER

The C-suite mandate

Underwriting modernization is not a departmental initiative. It is a capital allocation decision that reshapes risk posture. Executive priorities determine whether precision underwriting becomes reality.

Invest in AI-ready data for insurance, not isolated tools. AI-ready data is a signal to boards:

Sustainable AI performance depends on data foundations, not experimentation velocity.

AI-ready underwriting data means:

• Canonical submission models
• Structured and unstructured fusion
• Clear lineage and audit trails
• Embedded fairness testing
• Drift detection at the feature and outcome level

Without this, scaling AI increases systemic exposure.

DIVIDER

Shift from model governance to decision governance in insurance

Monitoring predictive accuracy alone is insufficient for effective underwriting governance.

Executives must monitor:

• Decision distributions
• Override behavior
• Referral rates
• Disparate impact signals
• Appetite alignment

This is decision operations. It is how insurers reduce hidden risk accumulation.

DIVIDER

Tie underwriting to continuous risk engagement

Telematics adoption continues to expand, with forecasts showing significant growth in insurance telematics markets and policy-level adoption rising in North America.

Connected sensor ecosystems and prevention-linked programs are shifting underwriting from a one-time gate into an ongoing risk partnership. Continuous underwriting is not branding language. It is the structural response to real-time risk signals.

DIVIDER

The strategic choice

Precision underwriting is not a technology upgrade. It is a structural redesign of how insurers perceive, evaluate, and monitor risk. Executives who treat AI as a productivity tool will capture incremental gains.

Executives who redesign underwriting as a governed, multimodal, continuously monitored decision system will reshape their risk economics. The difference will not be visible in a single quarter. It will be visible in who scales with confidence, who withstands regulatory scrutiny, and who adapts fastest when risk shifts.

DIVIDER

Conclusion

AI ambition is not the issue. Fragmentation is. When models scale faster than workflows and governance, performance stalls. Loss ratios don’t move. Advantage stays theoretical. The insurers pulling ahead are not running more pilots. They are redesigning how decisions flow across claims, underwriting, operations, and governance.

If you’re ready to move from experimentation to enterprise performance, start with our guide” The Insurance AI Flight Plan” — a practical blueprint for turning AI investment into measurable underwriting and claims impact.