Insurance AI Pivots From Claims Efficiency to Underwriting: What the ILTF 2026 Shift Means for Carriers

Having tracked AI deployment narratives across 15 carrier earnings calls this quarter, the shift from claims-efficiency language to underwriting-alpha framing is unmistakable. Evan Greenberg at Chubb talks about “agentics within AI” as a five-year growth catalyst. Christopher Swift at Hartford describes an “AI-first mindset” for reimagining workflows. Peter Zaffino at AIG details an orchestration layer coordinating multiple AI agents across the enterprise. The common thread: underwriting, not claims, is where carriers now expect AI to generate competitive separation.

Datos Insights crystallized this consensus at the 2026 Insurance Leaders Technology Forum (ILTF) in Boston on April 22-23. More than 100 carrier technology executives and 35 vendor sponsors gathered at the InterContinental under a banner theme they called “Defining the Intelligent Insurer.” The post-conference data tells a striking story: the pilot phase is over, production is scaling rapidly, and the strategic center of gravity for AI investment has moved decisively from claims processing to underwriting differentiation.

This article traces the strategic pivot through the ILTF survey data, the new “Intelligent Insurer Operating Model” framework, carrier earnings disclosures, and the reopening of the build-vs-buy debate that is forcing actuarial and technology leaders to rethink vendor strategy.

The ILTF Data: From Pilots to Production in 12 Months

The most consequential data point from the ILTF comes from a pre-conference survey of 36 senior carrier technology leaders. The share of carriers with AI in production grew from 37% to 61% in a single year. That 24-percentage-point jump represents a phase transition, not an incremental shift. Carriers are no longer evaluating AI; they are operationalizing it.

But production deployment does not mean uniform maturity. The survey reveals a paradox that should concern anyone building an AI strategy: only 8% of respondents believe they currently lead their peers in AI capabilities, yet 70% expect AI to deliver moderate competitive advantage within three years. That confidence-competence gap suggests many carriers are betting on AI’s future returns without having yet validated whether their current deployments generate measurable business impact.

Spending patterns reinforce this reading. Fully 70% of surveyed carriers spend under $500,000 annually on AI projects. For context, a single senior data scientist in New York costs $250,000-$350,000 in total compensation. These are not transformational budgets; they are experimentation budgets that happen to be running in production environments.

The production use case breakdown is equally revealing. Four of the top five deployed use cases fall within document processing: reading, extracting, summarizing, and classifying. These are important operational capabilities, but they are table-stakes automation, not competitive differentiation. Function-specific deployment tells a more nuanced story: underwriting AI is live at 56% of surveyed carriers, while claims AI sits at 50%. The six-point gap may look small, but it represents a directional reversal. Through 2024, claims processing consistently led underwriting in AI deployment priority across industry surveys.

Why Underwriting Overtook Claims

The pivot from claims to underwriting reflects three converging forces that became clear in the 2025 earnings cycle and sharpened at the ILTF.

Declining marginal returns on claims automation. Claims processing was the natural first target for AI because it offered clear, measurable efficiency gains: faster FNOL intake, automated damage estimation, straight-through processing for low-complexity claims. But after three to four years of investment, the easy wins have been captured. As the Roots AI 2026 predictions report notes, the real differentiator is no longer whether a carrier adopts AI but how it adopts it. Our analysis of the 82% adoption vs. 7% scalable success gap documented this pattern: high adoption rates mask thin deployment, and claims automation has hit the scaling wall first because it was deployed first.

Bigger pricing alpha from real-time risk selection. Underwriting offers something claims automation cannot: the ability to influence loss ratios at the point of risk acceptance rather than after a loss has occurred. When AI improves underwriting accuracy, the financial impact compounds across the entire policy lifecycle. Commercial P&C insurers implementing agentic AI systems report loss ratio improvements of 3-5 percentage points, according to industry benchmarks. A 4-point improvement on a $500 million premium book is $20 million in annual underwriting income, a return that dwarfs the marginal claims processing savings from the next incremental automation project.

Agentic AI unlocked underwriting complexity. The first wave of insurance AI relied on narrow models: computer vision for damage photos, NLP for intake classification, predictive models for fraud scoring. These tools worked well in claims because claims workflows are largely reactive and structured. Underwriting, by contrast, requires synthesizing unstructured data from submissions, loss runs, financial statements, regulatory filings, and broker commentary into a risk judgment. That synthesis task was beyond 2022-era AI. It is squarely within the capabilities of 2026-era agentic systems that coordinate multiple specialized models. SAS projects that underwriting will move from rule-based to relationship-based AI, with models learning from longitudinal customer data and recalibrating risk dynamically.

The Intelligent Insurer Operating Model

Datos Insights introduced the “Intelligent Insurer Operating Model” at the ILTF as a framework for how carriers should restructure around AI capabilities. The core proposition: replace traditional linear, siloed workflows with coordinated processes that combine human judgment and AI execution across the insurance value chain.

The framework represents Datos Insights’ attempt to move the industry conversation beyond “where should we deploy AI?” to “how should our operating model change to make AI effective?” The distinction matters. Most carrier AI deployments today bolt AI tools onto existing workflows. The Intelligent Insurer model argues for redesigning the workflows themselves, with AI as a structural component rather than an overlay.

Kurt Diederich, CEO of Finys, advanced a complementary argument in a Carrier Management executive viewpoint published the week after the ILTF. His framing: “The challenge has shifted from innovation to selection and, ultimately, to lifecycle management.” Carriers should treat AI as a modular capability within a “plug-and-play operating model,” enabling evaluation, implementation, and replacement of components with minimal disruption as conditions evolve.

The plug-and-play concept maps directly to the vendor landscape reality. The established AI use cases in insurance, including submission intake, underwriting support, claims triage, document processing, and customer service augmentation, now have multiple competing vendor solutions. Carriers that lock into a single vendor’s end-to-end platform risk accumulating what Diederich calls “technical and operational constraints that limit their abilities to adapt.”

The practical implication for actuaries: pricing models, reserve assumptions, and expense ratio projections built around specific vendor capabilities may need to be modular as well, with documented assumptions about which AI components are load-bearing and what happens to financial projections if a component is swapped or deprecated.

Quote-to-Bind Compression: The Metrics Behind the Pivot

The financial case for the underwriting pivot becomes concrete when you examine quote-to-bind cycle data. Industry benchmarks now show:

Metric	Pre-AI Baseline	AI-Enabled (2026)	Improvement
Straight-through processing rate	10-15%	70-90%	5-7x increase
Quote-to-bind cycle time	3 days (typical)	Minutes	Up to 99% reduction
Loss ratio impact	Baseline	3-5 point improvement	Direct underwriting income
P&C underwriting expense ratio	Baseline	15-20% decline projected	Structural cost reduction
Life underwriting expense ratio	Baseline	25%+ decline projected	Larger automation scope

The straight-through processing (STP) leap from 10-15% to 70-90% is the most structurally significant number in this table. An STP transaction requires no human intervention from submission to policy issuance. When STP rates were 10-15%, AI’s role in underwriting was marginal: handling the simplest risks while humans processed everything else. At 70-90%, the dynamic inverts. Humans focus on the complex, high-value risks that require judgment, while AI handles the volume. That is a fundamentally different operating model, precisely what the Intelligent Insurer framework envisions.

These figures are not theoretical projections. AIG’s Lexington Insurance unit processed more than 370,000 submissions in 2025 using generative AI, with a target of 500,000 by 2030. The 55% time-to-quote reduction and 40% binding lift reported across AIG Assist’s eight commercial lines represent production metrics, not pilot results.

Three Carrier Models for the AI-First Insurer

The ILTF survey data and Q1 2026 earnings calls reveal distinct strategic approaches among leading carriers. These are not variations on a theme; they represent fundamentally different bets on how AI will create value in insurance.

Chubb: Disciplined Incrementalism

Evan Greenberg’s approach at Chubb is unmistakable from his Q1 2026 commentary: broad-based investment in “agentics within AI” and “evolving large language model capabilities,” but with a conspicuous emphasis on executive knowledge. “You got to have firsthand knowledge. You can’t just be listening to others,” Greenberg told analysts, signaling that Chubb’s leadership is personally evaluating AI capabilities rather than delegating entirely to technology teams.

Chubb’s digital transformation remains “on track” and “steady,” with AI positioned as a five-year growth catalyst for small commercial and E&S segments, both in North America and internationally. The strategy prioritizes operational efficiency and intermediation cost reduction over transformational restructuring. Chubb is not attempting to reinvent its operating model; it is systematically embedding AI into an operating model that already produces an 84% combined ratio.

The actuarial implication: when your underwriting results are already industry-leading, the risk calculus around AI deployment shifts. The downside of a failed AI initiative is larger because there is more to protect. Chubb’s incremental approach reflects that calculus.

Hartford: AI-First Workflow Reimagination

Christopher Swift’s framing at Hartford is more ambitious. The company has “moved to the next phase of our innovation agenda, reimagining our processes and workflows with an AI-first mindset,” according to the Q4 2025 earnings wrap. Hartford is deploying AI across three domains simultaneously: claims record summarization, underwriting data-rich insights, and operations via Amazon contact center technology.

Hartford’s cloud-based Prevail platform for auto and home insurance was extended to agents in 10 states during 2025, with expansion planned to 30 states by 2027. Swift expressed confidence that borders on bullish: “For us, really, the sky is the limit.” The strategic emphasis on “practical, high-impact applications that augment human talent” suggests Hartford is pursuing the middle path, not as conservative as Chubb’s incrementalism, but not as architecturally ambitious as AIG’s multi-agent system.

Hartford’s Q1 2026 earnings reinforce the approach: Business Insurance delivered 6% written premium growth with an underlying combined ratio of 89.2%, with an AI assistant augmenting underwriting workflows in an increasingly competitive environment. The personal lines turnaround is being tested in a live market, with AI-augmented workflows contributing to underwriting discipline as rates earn through.

AIG: Multi-Agent Orchestration

AIG represents the most architecturally aggressive approach. The company has deployed an orchestration layer designed to coordinate multiple AI agents across the enterprise. CEO Peter Zaffino described these agents as “companions that operate with our teams,” providing real-time information delivery, historical case analysis and reference, and underwriting decision challenges and validation.

“We orchestrate agents so they can scale and be able to analyse that information” without bias throughout workflows, Zaffino told investors. The orchestration layer determines when agents are activated, how they share information, and when human oversight is required.

AIG’s internal platform, AIG Assist, is now implemented across most commercial lines of business. The March 2026 McGill and Partners collaboration, which targets up to $1.6 billion of specialty gross premiums written via agentic AI and Palantir’s Foundry platform, signals AIG’s willingness to extend multi-agent architecture beyond internal operations into the broker distribution channel.

The three models create distinct risk profiles that actuaries should map to their own organizations:

Carrier	AI Strategy	Primary Target	Risk Profile
Chubb	Disciplined incrementalism	Small commercial, E&S growth	Low disruption, slower upside capture
Hartford	AI-first workflow redesign	Underwriting + claims + operations	Moderate; depends on platform execution
AIG	Multi-agent orchestration	Enterprise-wide, including distribution	Highest upside, highest complexity risk

The Build-vs-Buy Question Reopens

One of the most consequential findings from the ILTF was the reopening of the build-vs-buy debate. AI has compressed development timelines to the point where building proprietary capabilities is competitive again, a reversal from the 2020-2024 period when most carriers defaulted to vendor solutions because internal development was too slow and too expensive.

The shift tracks with broader market dynamics. Deloitte, Oliver Wyman, and McKinsey all flagged the tension between vendor dependency and competitive differentiation in their 2026 insurance outlooks. Our analysis of Guidewire’s PricingCenter examined this calculus specifically for actuarial teams, where the build-vs-buy decision directly affects model transparency, validation workflows, and regulatory documentation requirements.

Diederich’s Carrier Management piece offers a cautionary historical parallel. The current insurtech vendor landscape, characterized by “high entrant volume, uneven differentiation, and ongoing capability leapfrogging,” mirrors the early 2000s internet proliferation, “where only a small percentage of vendors ultimately proved durable.” Carriers that invested heavily in Pets.com-era internet strategies learned painful lessons about vendor selection. The AI vendor landscape in insurance is heading for a similar consolidation cycle.

The ILTF data supports this concern. With 70% of carriers spending under $500,000 annually on AI, most are relying heavily on vendor solutions rather than building internal capabilities. That creates concentration risk: if a key vendor is acquired, pivots its product strategy, or fails, the carrier’s AI-dependent workflows break. The plug-and-play model that Diederich advocates is partly a risk management strategy for this exact scenario.

For actuaries, the build-vs-buy decision has direct implications for model governance. Vendor-supplied AI models often operate as partial black boxes, with limited access to training data, feature weights, or drift monitoring. Building internally offers greater transparency, but requires hiring and retaining data science talent in a market where converting AI spend into measurable actuarial results remains the exception. The 8% of ILTF respondents who believe they lead peers in AI capabilities are likely the carriers that have resolved this tension; the other 92% are still navigating it.

Consumer Sentiment: The Constraint Nobody Discussed

One dimension conspicuously absent from the ILTF’s underwriting-centric narrative: consumer acceptance. An April 2026 Insurity survey of over 1,000 U.S. adults found that P&C consumer support for AI nearly doubled in a single year, from 20% to 39%. That sounds encouraging until you examine the specifics.

Consumer comfort drops sharply when AI moves from support functions to decision-making. Only 22% are comfortable with AI filing a claim on their behalf. Just 16% accept AI canceling or renewing a policy. The pattern is consistent: consumers support AI as an efficiency tool for their insurer, but resist AI making consequential decisions about their coverage.

This creates a practical ceiling on how far the underwriting pivot can go in personal lines. Commercial lines, where the “customer” is a risk manager or broker who evaluates AI capabilities as a purchasing criterion, face fewer acceptance constraints. That distinction helps explain why the most aggressive AI underwriting deployments, including AIG’s Lexington unit, Hartford’s AI-assisted workflows, and the ILTF’s featured case studies, cluster in commercial and specialty segments.

Why This Matters for Actuaries

The claims-to-underwriting pivot has specific, tangible implications for actuarial practice across pricing, reserving, and enterprise risk management.

Expense ratio assumptions need revisiting. If underwriting expense ratios decline 15-20% in P&C as projected, that changes rate level indications, competitive positioning analysis, and long-term profit projections. Pricing actuaries should be building scenario models around AI-driven expense reduction, not as a hypothetical but as a near-term planning assumption for renewal years.

Loss ratio volatility may shift. AI-improved risk selection should reduce adverse selection and improve loss ratios at the book level. But it may also create selection risk in segments where one carrier has materially better AI than competitors, winning the better risks and leaving adverse portfolios for carriers with weaker selection capabilities. This is the classic winner’s curse dynamic, accelerated by technology.

Reserve actuaries face new model governance questions. When AI agents influence underwriting decisions, the boundary between “underwriting model” and “pricing model” blurs. Reserve actuaries need to understand whether AI-influenced portfolio composition changes are reflected in actuarial assumptions. The governance gap we documented in March becomes more acute as AI moves from claims (where it affects costs) to underwriting (where it affects the distribution of risks entering the book).

The Intelligent Insurer Operating Model demands actuarial involvement in design. If carriers restructure workflows around coordinated human-AI processes, actuaries need to be in the design room, not brought in after the architecture is set. The financial feedback loops that actuaries understand, including how underwriting decisions create reserve liabilities that affect capital allocation that constrains future underwriting capacity, are precisely the dynamics that need to be encoded into any operating model redesign.

Vendor risk becomes an actuarial consideration. The build-vs-buy tension, combined with the vendor consolidation risk that Diederich highlighted, means that actuaries preparing opinions on reserves or capital adequacy may need to consider the operational risk of AI vendor dependency. If a carrier’s 70-90% STP rate depends on a third-party AI platform, what happens to expense ratios and processing capacity if that platform experiences an outage or contract termination? These scenarios belong in enterprise risk assessments.

The P&C combined ratio improved to approximately 95% in 2025, the strongest underwriting performance in roughly a decade. Premium growth is projected at 4% for 2026. In a market where underwriting results are good and growth is decelerating, the carriers that pull ahead will be those that use AI to maintain underwriting discipline while finding profitable growth in segments their competitors cannot efficiently access. That is the strategic logic behind the underwriting pivot, and it places AI squarely in the actuarial domain.