Why Carriers Deploy Agentic AI in E&S Lines Before All Else

Every major carrier agentic AI deployment announced in the past 18 months shares one feature: it started in excess and surplus lines. AIG built its multi-agent underwriting stack around Lexington Insurance, its E&S subsidiary. Ki Insurance launched the first fully algorithmic Lloyd’s syndicate writing specialty and surplus business. Kinsale Capital, a pure-play E&S carrier, has woven AI agents throughout its proprietary operating system. Hiscox compressed quote turnaround from three days to three minutes in its London Market specialty book.

This is not coincidence. From tracking these deployments over the past two years, a clear pattern has emerged: E&S lines offer a combination of regulatory freedom, submission volume, processing inefficiency, and profitability headroom that no admitted market can replicate. Carriers are not choosing E&S for pilot programs out of convenience. They are choosing it because the economics and regulatory structure make E&S the only segment where agentic AI can be deployed at production scale without waiting years for state-by-state approval.

This article maps the structural reasons behind that choice, drawing on AM Best premium data, carrier earnings disclosures, NAIC regulatory frameworks, and published deployment metrics from six carriers and three technology platforms.

The $129.8 Billion Beachhead

The U.S. surplus lines market has grown from a niche corner of commercial insurance into a segment that now accounts for 12.3% of total P&C direct premiums written. AM Best reported that E&S DPW reached $129.8 billion in 2024, marking the seventh consecutive year of double-digit growth at 12.3% year-over-year. As recently as 2000, surplus lines represented just 3.6% of the broader P&C market.

The growth trajectory has been striking across specific lines. E&S homeowners premium surged 29.5% to $4.14 billion in 2025, the third consecutive year of 20%-plus increases, driven largely by admitted carriers withdrawing from catastrophe-exposed geographies. Commercial property, by contrast, declined 2.8% to $27.69 billion as competition intensified on larger accounts.

Through the first three quarters of 2025, overall E&S premium expansion moderated to 9.7%, and AM Best revised its segment outlook from positive to stable in late 2025. Fitch Ratings projects E&S market share will peak around 12% before the cycle turns. But even with softening momentum, the market’s absolute size creates a massive opportunity for AI-driven efficiency gains.

The concentration dynamics matter for understanding AI deployment incentives. The top 25 groups plus Lloyd’s accounted for roughly two-thirds of the $129.8 billion total in 2024. These are the same carriers with the technology budgets, data infrastructure, and strategic ambition to invest in agentic AI. The market is large enough to justify the investment and concentrated enough that a handful of early movers can reshape competitive dynamics.

Rate-Filing Freedom: The Structural Advantage

The single most important reason carriers deploy AI in E&S first is regulatory. In admitted markets, insurers must file rates and policy forms with each state’s department of insurance before using them. Depending on the state, this means prior approval, file-and-use, or use-and-file systems, each with its own timeline and documentation requirements. An AI-driven pricing model deployed in an admitted line faces the same regulatory gauntlet as any other rate change: actuarial memoranda, supporting data, bias testing documentation, and potentially months of examiner review.

Surplus lines carriers face none of this. As the Wholesale & Specialty Insurance Association (WSIA) states plainly: in a surplus lines transaction, the insurer can use “whatever rates and policy provisions the insurer and applicant agree upon.” The insurer is not subject to the home state’s rate and form filing regulations. This means a carrier can deploy an agentic AI pricing model, observe its performance, adjust the model, and redeploy, all without filing a single document with a state regulator.

The timeline differential is substantial. Milliman data shows that Colorado personal auto rate filings averaged 373 days from initial filing to approval. Some states take five to six months after initial review batches. Prior approval states carry the second-highest combined ratios among all regulatory frameworks, reflecting the friction between market conditions and the pace of regulatory response.

For AI models specifically, the documentation burden in admitted markets is growing. State examiners now require data lineage documentation covering source systems, extraction dates, sample construction rules, and missing value treatment. They demand feature-to-factor mapping showing how a model’s 47 technical inputs map to 12 filed rating factors with actuarial justification for each. Bias testing evidence must go beyond attestations to include protected classes tested, proxy variables evaluated, statistical methods, thresholds, and remediation results. As one regulatory analysis noted, examiners have explicitly rejected SHAP plots alone: “A SHAP plot is not an explanation. It is a graph.”

In E&S, a carrier can iterate on its AI models continuously. It can test new data sources, adjust risk appetite algorithms, and refine pricing in response to observed loss emergence, all without waiting for regulatory clearance. This creates an R&D velocity advantage that compounds over time. Every month an E&S carrier runs its AI models in production generates training data and performance feedback that admitted-market competitors simply cannot access at the same pace.

The Submission Crisis: 50% of Volume Never Gets Underwritten

The E&S market operates on a fundamentally different distribution model than admitted lines. Risks flow from retail agents to wholesale brokers to surplus lines carriers, generating massive submission volumes at each stage. AIG’s Lexington Insurance alone surpassed 370,000 submissions in 2025 and is targeting 500,000 by 2030. Across the broader market, the numbers are staggering.

And most of those submissions never receive a proper review. Industry data indicates that roughly 50% of submissions, whether outside appetite or arriving incomplete, never get properly underwritten. Only 30-40% of submissions receive a thorough review at most carriers. The remaining volume either sits in queue until it expires, receives a cursory declination, or gets lost entirely in email inboxes and shared drives.

The root cause is a structural mismatch between volume and capacity. Senior underwriters spend 60-70% of their time on data extraction and administrative tasks rather than risk evaluation. An estimated 40% of underwriter time goes to work that does not require underwriting expertise: pulling data from submissions, reformatting it into internal systems, and chasing brokers for missing information.

The talent crisis amplifies the problem. The insurance industry expects to lose approximately 400,000 professionals by 2026, with specialized underwriting positions remaining unfilled for an average of 4.2 months. Nearly 35% of insurance professionals are approaching retirement age, and one in three young professionals are considering leaving the industry. The actuarial profession faces a projected 25% shortage over the next five years.

In soft markets, the submission problem compounds dramatically. Accounts that generated two submissions in hard markets may generate five or six submissions as brokers shop more aggressively, without proportional staffing increases at carriers. Swiss Re forecasts premium growth slowing from 9.6% in 2024 to 3% in 2026, with combined ratios expected to worsen from 97.2% to 99%. The economics of hiring more underwriters to handle cyclical volume spikes simply do not work.

This is precisely the environment where agentic AI delivers its highest ROI. An AI system that can triage submissions, extract data from unstructured documents, and generate preliminary risk assessments does not solve a convenience problem. It solves a fundamental capacity problem that determines how much premium a carrier can write.

Speed-to-Quote: From Days to Seconds

The published cycle-time reductions from early E&S AI deployments are not incremental improvements. They represent order-of-magnitude changes in how fast carriers can respond to submission flow.

Hiscox deployed an agentic AI system for its Sabotage & Terrorism line, built on Google Gemini and Vertex AI with Hailo. The system reads incoming email submissions, extracts 15-plus data points, cleanses and geocodes addresses, and produces a structured risk profile autonomously. Quote turnaround dropped from three days to three minutes, a 99.4% cycle time reduction.

Ki Insurance, the world’s first fully algorithmic Lloyd’s syndicate (Syndicate 1618), generates follow lines in as little as 10 seconds compared to the two-to-three-week traditional lead time for Lloyd’s placement. Ki has grown to $1.11 billion in managed premium in 2025, backed by $500 million in initial capital from Blackstone and Fairfax, with capacity partners including Beazley, QBE, Aspen, Travelers, and TMK. Its 50-plus ML models powered an 89.7% combined ratio on $877 million GWP in 2023, a seven-point improvement from the prior year.

Pathpoint, a digital E&S platform, delivers bindable quotes for 80% of submissions instantly, typically in under 60 seconds. In California, 75.9% of submissions return an instant quote in approximately 14 seconds. The platform connects to 27-plus AM Best A-rated carriers across multiple E&S lines.

AIG Lexington achieved a 55% reduction in time to quote for underwriters using its AIG Assist platform, alongside a 30% improvement on quoting submissions and a roughly 40% increase in binding of submissions. CEO Peter Zaffino described the impact: “We’re seeing a massive change in our ability to process a submission flow way [without] additional human capital resources.”

Allianz compressed decision times from three to five days to approximately 12 minutes while maintaining 99%-plus accuracy. N2G Worldwide reported a 60% reduction in cycle times with a 40% increase in underwriter quote capacity.

These speed advantages translate directly to profitability. Carriers using automated systems report 100-200% increases in underwriting efficiency. A three-to-five percentage point loss ratio improvement, documented across multiple implementations, translates to $40 million in annual profit on a $1 billion premium portfolio. New business premium increases of 10-15% and broker retention gains of 5-10% have been measured at carriers with sub-60-second quote delivery.

From tracking these metrics across carriers, one pattern stands out: the speed advantage matters most in E&S precisely because of the submission-to-bind funnel dynamics. When only 50% of submissions get reviewed and only 30% of those generate quotes, the carrier that quotes first wins a disproportionate share of the bindable business. Speed is not just operational efficiency; it is a selection advantage that improves portfolio quality by accessing risks before competitors can respond.

Carrier Deployments: Who Is Already There

The carrier case studies are no longer theoretical. Several E&S-focused insurers have moved from pilot programs to production-scale agentic AI deployments.

AIG has deployed AIG Assist across eight lines of business for core underwriting and claims. The platform uses a multi-agentic orchestration layer that coordinates specialized AI agents for risk evaluation against underwriting guidelines and pricing benchmarking against portfolio targets. AIG runs this stack on Palantir’s Foundry platform with Anthropic’s Claude as the underlying LLM. In March 2026, AIG extended its agentic AI approach to Lloyd’s through a partnership with McGill and Partners to underwrite up to $1.6 billion of specialty GWP using Palantir-powered algorithmic follow underwriting.

Kinsale Capital, a pure-play E&S carrier, has built AI into its proprietary operating system from the ground up. Unlike carriers retrofitting AI onto legacy policy administration systems, Kinsale operates on a custom-built core platform with no legacy software constraints. The results are measurable: underwriters process 24.5% more quotes daily than before AI adoption, freeing up approximately 10 hours per week per underwriter from repetitive tasks. CEO Michael Kehoe describes technology as a “core competency” with “extensive use of AI models” and “various AI agents” embedded in the enterprise system. Kinsale’s structural cost advantage, estimated at roughly 15 points below peer expense ratios, continues to widen through technology investment.

Ki Insurance separated from Brit as a standalone company in January 2025, validating the fully algorithmic underwriting model. Its $1.11 billion in managed premium includes $200 million-plus of partner capacity, demonstrating that traditional carriers are willing to allocate capital behind algorithmic risk selection. Ki was built on Google Cloud with an academic partnership with University College London, and its 50-plus ML models provide a technical depth that most carrier AI programs have not yet matched.

James River Group selected Kalepa’s AI platform to enhance its E&S underwriting. The platform automates submission intake, triages risks, and centralizes risk-critical information from various sources to deliver decision-ready quotes. This represents the vendor-adoption path: mid-market E&S carriers that lack AIG or Kinsale’s internal engineering capacity can still deploy agentic AI through specialized insurtechs.

Ryan Specialty, the largest E&S-focused intermediary, has committed to accelerating AI deployment through its RT Connector platform, which now offers real-time bindable quotes and instant policy issuance across 25-plus carriers and eight lines of business. Ryan’s leadership has stated the firm believes it is a “net beneficiary” of AI-driven transformation in wholesale and delegated authority.

The broader market data confirms the trend. An April 2025 London Market Association survey of 81 firms found that 33% were actively using AI tools, 47% were experimenting without wide adoption, and 14% had deployed or experimented with agentic or generative AI specifically in underwriting. The dominant use case, at 74%, was data extraction from unstructured documents, followed by submission preparation at 54%.

Profitability Headroom: Why the Math Works in E&S

E&S carriers operate at a significant profitability advantage over the broader P&C market, and this headroom makes AI investment economics more favorable.

In 2024, the E&S direct combined ratio was 88% compared to 95% for the total P&C market, a seven-point advantage. The DPSL loss and LAE ratio of 63.3 was 7.8 points better than the overall P&C industry. While this gap narrowed from 14.8 points in 2023 to 7.8 points in 2024, E&S remains consistently more profitable than admitted lines.

This profitability headroom matters for AI investment in two ways. First, it provides the earnings cushion to fund technology investment without jeopardizing underwriting results. Second, it creates a wider margin of error for AI-driven pricing decisions during the learning period when models are still calibrating against actual loss emergence.

Ki Insurance demonstrates what algorithmic underwriting profitability looks like at scale. Its 2023 combined ratio of 89.7% on $877 million GWP produced $101 million in profit, with a seven-point combined ratio improvement from 2022. This result was achieved with a fully algorithmic approach, not a human-augmented one.

General liability rates in E&S continue to support pricing adequacy, with increases ranging from 3.95% to 7.23% on primary GL and 9.49% on umbrella liability. These rate levels provide margin for carriers to absorb the technology investment cost while maintaining underwriting discipline.

The Admitted Market Barrier Wall

If E&S offers a regulatory greenfield for AI deployment, admitted lines present the opposite: a thicket of filing requirements, documentation mandates, and emerging AI-specific regulations that add months or years to deployment timelines.

The NAIC AI Model Bulletin, adopted in December 2023, has been adopted by over half of all states as of early 2026. It requires governance and risk management controls, oversight of third-party AI vendors, and records of data sources, testing, bias analysis, and model drift monitoring. While the bulletin is principle-based rather than prescriptive, it creates a documentation burden that surplus lines carriers simply do not face.

Colorado’s SB 21-169 goes further, requiring insurers to ensure AI systems are explainable and auditable, with documentation of data sources, model design, training, testing, and monitoring. Bias audits and fairness assessments are mandatory. The amended regulation became effective October 15, 2025, with life insurer compliance attestations due by December 1, 2024.

For carriers using AI in admitted-market rate filings, state examiners have raised the bar substantially. Vague statements like “five years of policy data” are insufficient; examiners require exact source systems, extraction dates, and sample construction rules. Model inventories must include version numbers, training dates, validation dates, and ownership documentation. Vendor documentation sufficiency standards and ongoing monitoring commitments are evaluated during the filing review.

The practical effect is a years-long timeline gap. A carrier can deploy an agentic AI underwriting model in its E&S book today, measure its performance over 12-18 months of actual loss emergence, and then use that track record to support the documentation and validation requirements for admitted-market filings. E&S serves as the production proving ground; admitted markets follow once the evidence base exists.

Wholesale Broker APIs and the Distribution Plumbing

E&S distribution creates additional conditions favorable to AI deployment. The wholesale broker channel that connects retail agents to surplus lines carriers is rapidly building API infrastructure that enables automated submission routing, real-time quoting, and instant binding.

The three largest wholesale distributors are investing heavily in connectivity. Amwins, the largest independent wholesale distributor with 7,300 employees and 155-plus offices globally, handles $33 billion-plus in annual premium placements and integrates with policy management platforms via APIs. Ryan Specialty’s RT Connector offers real-time quotes across 25-plus carriers and eight lines. CRC Group serves the most difficult 10% of risks and has been building digital submission capabilities.

These intermediaries solve a two-sided marketplace problem: connecting roughly 60,000 retail agents in the U.S. to approximately 1,000 carriers across the U.S., Bermuda, Europe, and London. API connectivity enables agentic AI at the carrier level to receive structured submission data rather than parsing emailed PDFs, dramatically improving the speed and accuracy of automated underwriting.

Compliance automation reinforces the trend. The Surplus Lines Automation Suite (SLAS) has been adopted by 11 states and processes more than one-third of U.S. surplus lines premium annually. InsCipher’s platform covers approximately 90% of surplus lines filing volume nationally across 19 integrated states, with API integrations that can supercharge filing team velocity by 300%. As this compliance infrastructure matures, it removes one of the remaining manual bottlenecks in the E&S workflow.

Why This Matters for Actuaries

The E&S-first deployment pattern has direct implications for actuarial practice across several dimensions.

Model validation is shifting. Actuaries validating AI underwriting models in E&S will encounter different governance requirements than those working on admitted-market filings. The absence of rate-filing documentation does not mean the absence of actuarial standards. ASOP No. 56 still governs modeling practices, and ASOP No. 23 governs data quality. The challenge is applying these standards to agentic systems where multiple AI agents coordinate to produce an underwriting decision, and where the model evolves continuously rather than being filed at fixed intervals.

Loss emergence data from E&S AI deployments will inform admitted-market pricing. As carriers accumulate loss experience from AI-underwritten E&S books, that data will flow into the actuarial basis for admitted-market rate filings. Actuaries will need to evaluate whether E&S loss experience is credible and applicable to admitted risks, considering differences in policy terms, risk selection, and distribution channels.

Career opportunities are concentrating in E&S-focused carriers. With 22% of insurers planning agentic AI in production by year-end 2026 and current AI adoption in underwriting at just 14%, the talent gap is significant. Actuaries who understand both traditional ratemaking and AI model governance are positioned at the intersection of the two capabilities carriers need most. The projected 70% adoption rate for AI in underwriting by 2028 suggests this demand will intensify.

Reserve adequacy analysis is evolving. AI-underwritten books may exhibit different loss development patterns than traditionally underwritten portfolios. Changes in risk selection, speed of binding, and the mix of business written through AI versus human underwriters all affect loss emergence timing and ultimate loss ratios. Appointed actuaries responsible for statutory opinions will need to develop methods for segmenting and analyzing AI-influenced reserves separately.