Markel AI Strategy: How a Specialty Insurer Modernizes Complex Underwriting With Conversational AI

TL;DR#

Markel Group (NYSE: MKL) is a diversified specialty insurer whose ai insurance opportunity sits squarely in complex, hard-to-place underwriting — the excess and surplus (E&S) lines where admitted carriers won't go. Markel is the sixth-largest E&S writer in the United States, carries an A (Excellent) rating from AM Best, and reported $9.4 billion in gross written premium for 2024 inside a three-engine model spanning Insurance, Investments, and Markel Ventures. The company has already moved on AI: it was the first insurer globally to deploy Microsoft 365 Copilot to all staff (passing 60% active usage), put legal-reading tools like Harvey into its M&A underwriting, and partnered with Cytora to turn broker submissions into structured digital risk flows. The strategic gap most specialty carriers — including Markel — still face is the front door: the messy, conversational intake where a broker or insured describes a non-standard risk that no ACORD form fully captures. Conversational AI that interviews the submitter, probes ambiguous exposures, and structures the answer is the natural next layer on top of the document-extraction tooling Markel already runs. This analysis is grounded in public disclosures; where Markel has not announced a specific deployment, it is framed as an opportunity, not a claim.

Markel's Profile: Why Specialty Underwriting Is the Hard Case for AI#

Markel is a specialty insurer built to underwrite the risks standard carriers reject, which makes it one of the most demanding — and most valuable — environments for ai insurance to prove itself. Unlike a personal-auto or homeowners carrier underwriting millions of near-identical policies, Markel concentrates on excess and surplus lines: companies with unusual risk profiles, very high hazard, extensive claims histories, or operations in new and emerging industries. The E&S segment is often called the "safety valve" of the insurance industry precisely because it absorbs what the admitted market can't price.

That concentration shows up in the numbers. Markel's Insurance segment generated $8.7 billion in operating revenue and $601.0 million in operating income in 2024, with the combined ratio improving to 94.3% from 97.8% a year earlier, according to Markel Group's 2024 financial results. The broader business is a three-engine holding company — Insurance, Investments, and the 21 operating companies inside Markel Ventures — sitting on $34.2 billion of invested assets at year-end 2024. CEO Simon Wilson has publicly recommitted Markel Insurance to its "bread-and-butter" E&S business written through wholesale brokers, as reported by Insurance Journal.

Why this matters for AI is the nature of the decision. A specialty submission is not a form — it's a narrative. The exposure lives in attachments, supplementals, loss runs, broker emails, and the gaps between them: the worst possible fit for static intake and the best possible fit for a conversation.

The Specialty E&S Context: Why Complex Underwriting Is Growing#

The complex-risk problem Markel underwrites is getting bigger, raising the stakes on every workflow it can automate. 2024 marked the seventh straight year of double-digit growth for U.S. excess and surplus lines, with direct premiums written reaching roughly $98.2 billion, up from $86.6 billion in 2023, according to AM Best data reported by Insurance Journal. Growth cooled slightly to 13.4%, but the trajectory is clear: more risk migrates into the specialty market every year.

That migration is reshaping who underwrites it. The U.S. managing general agent (MGA) market wrote an estimated $114.1 billion in direct premiums in 2024, growing 16% year over year as underwriting talent moves from carriers to specialty platforms — a more fragmented, broker-driven, document-heavy submission ecosystem where intake quality determines underwriting economics.

For a specialty carrier, the bottleneck isn't pricing sophistication; Markel already has actuarial models, catastrophe models, and IoT data feeds. The bottleneck is getting clean, complete, structured information about a non-standard risk into the underwriting workbench fast enough to quote before a competitor does. Our breakdown of commercial insurance AI for brokers, MGAs, and carriers covers how the submission funnel became the contested ground.

What Markel Has Actually Done With AI#

Markel has made real, documented AI investments — mostly in productivity and document processing rather than customer-facing conversation:

• Microsoft 365 Copilot, firm-wide. Markel was the first insurance company globally to roll out Microsoft 365 Copilot to all staff and reported over 60% active usage with early productivity gains, according to an interview with CEO Simon Wilson published by Sollers.
• Harvey for M&A legal review. In its M&A insurance business, Markel deployed Harvey, an AI system that reads legal paperwork in under 10 minutes versus roughly a week of underwriter time, per the same interview.
• Cytora for digital risk flows. Markel partnered with Cytora to turn unstructured broker submissions into structured, digital risk flows that route to the right underwriter, as described on Cytora's platform.

The pattern is telling. Each tackles the document problem — processing information that already exists in a file. Wilson has framed the broader ambition in racing terms, arguing that without market-leading operations and technology, competing is "like trying to race a bike against a Formula 1 car."

What none of these deployments do is talk to the submitter. They read what was sent; they don't interrogate what was missing. That is the open lane. Carriers across the market are testing the same boundary — see how a top-five carrier is approaching it in our look at Liberty Mutual's AI strategy and how a $260B specialty leader is positioning in our Chubb AI strategy analysis.

The Gap: Conversational AI at the Front Door of Underwriting#

The biggest untapped ai insurance opportunity for a specialty carrier like Markel is conversational intake — an AI that interviews the broker or insured to fill the holes document extraction leaves behind. Submission automation platforms already read broker emails, ACORD forms, loss runs, and statement-of-values schedules, with the best reporting up to 70% faster processing. But extraction has a ceiling: it can only structure what the submitter chose to include.

Specialty risk is defined by what's not in the file. A restaurant with a wood-fired oven, a contractor expanding into a new trade, a manufacturer that quietly added a hazardous process line — these exposures determine whether a risk is profitable, and they're routinely under-described in a first submission. Today an underwriter resolves them by firing an email chain back through the wholesale broker, burning days of cycle time on every account.

A conversational AI agent collapses that loop. Instead of a static supplemental form, the submitter has a guided conversation that adapts to their answers, probes vague descriptions ("you said 'light manufacturing' — what's actually being made, and with what equipment?"), and asks only the questions a given risk class requires. Forms flatten people into dropdowns, while conversational intake captures the "why" behind a risk. On the mechanics of replacing static forms, our piece on why static intake forms are killing your conversion rate maps onto the broker drop-off problem.

This is the layer Perspective AI's intelligent intake is built for — an AI interviewer agent that runs the structured-but-conversational risk interview, and a concierge agent that replaces the static supplemental form.

How Conversational Underwriting Intake Would Work for a Specialty Carrier#

Conversational underwriting intake works by replacing the back-and-forth email loop with one adaptive interview that produces structured, underwriting-ready data. For a carrier with Markel's profile, a realistic workflow looks like this:

1. Step 1 — Submission lands. A wholesale broker submits via email or portal; document-extraction tooling (the Cytora layer Markel already runs) parses the ACORD forms, loss runs, and attachments.
2. Step 2 — Gap detection. The system flags what's missing or ambiguous for that risk class — the fields an underwriter would otherwise chase.
3. Step 3 — Conversational fill. Instead of a generic supplemental, the broker or insured gets an AI-led interview that asks only the open questions, probes vague answers in plain language, and captures context a checkbox can't.
4. Step 4 — Structured handoff. The conversation is mapped into the underwriting workbench as clean fields plus a narrative, with a full transcript for the audit trail.
5. Step 5 — Underwriter decision. The underwriter opens a complete, decision-ready file instead of a partial one — quoting faster without sacrificing the judgment specialty risk demands.

The audit trail matters more in E&S than almost anywhere. The NAIC Model Bulletin on the use of AI by insurers, adopted by roughly two dozen states, requires documented governance for AI used in underwriting — making a transcript-backed, explainable intake conversation an asset, not a liability. The same principle drives the conversational FNOL shift in claims and the move toward conversational underwriting in life insurance, where 90-page applications are giving way to guided interviews.

Specialty Peers Show the Same Pattern#

Markel is not alone — the entire specialty and commercial cohort is converging on conversational risk intake, the clearest signal that the opportunity is real. Workers' comp specialist Pie Insurance has built an AI-first underwriting model, covered in our Pie Insurance workers' comp case study. Behavior-based pricing pioneer Root is making a similar bet, detailed in our analysis of Root's AI underwriting and the conversational risk interview. On the commercial side, AIG is testing conversational underwriting in commercial insurance, and regional carriers like Selective are betting on conversational risk intake. For the tooling landscape, our comparison of AI underwriting software by use case maps where each category fits.

The lesson is consistent: document extraction gets you to a partial file; conversation gets you to a complete one.

Frequently Asked Questions#

What is Markel's AI strategy?#

Markel's publicly documented AI strategy centers on productivity and document processing rather than customer-facing automation. It was the first insurer globally to deploy Microsoft 365 Copilot to all staff (over 60% active usage), uses Harvey to read legal paperwork in its M&A unit, and partnered with Cytora to convert broker submissions into digital risk flows. Markel has not announced a customer-facing conversational AI product, leaving conversational underwriting intake an open opportunity.

Why is specialty insurance harder to automate than personal lines?#

Specialty insurance is harder to automate because each risk is non-standard and described in narrative, not structured form fields. Excess and surplus lines cover companies with unusual profiles, high hazard, or operations in emerging industries, so critical exposure detail often lives in attachments, broker emails, and the gaps between documents — making static forms a poor fit and adaptive, conversational intake a much stronger one.

What is the difference between document extraction and conversational AI in underwriting?#

Document extraction reads and structures information that already exists in a submission, while conversational AI gathers the information that is missing. Extraction tools parse ACORD forms, loss runs, and emails; conversational agents interview the broker or insured to probe vague descriptions and fill gaps a checkbox can't capture. The two are complementary — extraction handles the file, conversation completes it.

How big is the excess and surplus lines market Markel competes in?#

The U.S. excess and surplus lines market reached roughly $98.2 billion in direct premiums written in 2024, up from $86.6 billion in 2023, marking a seventh consecutive year of double-digit growth. Markel is the sixth-largest E&S writer in the United States. The parallel MGA market wrote an estimated $114.1 billion in 2024, reflecting how much complex risk is migrating into the specialty channel.

Does conversational AI intake meet insurance compliance requirements?#

Conversational AI intake can support compliance when it produces a documented, explainable record. The NAIC Model Bulletin on AI, adopted by roughly two dozen states, requires governance and auditability for AI used in underwriting. A conversational agent that captures a full transcript and maps answers to structured fields creates exactly the audit trail regulators expect, often more transparently than a static form.

Conclusion: The Conversational Layer Is Markel's Open Lane#

Markel has already proven it will invest in ai insurance — firm-wide Copilot, Harvey in M&A, and Cytora for digital risk flows are real, documented moves that put it ahead of most specialty peers on the document side of underwriting. The unfinished work is the front door: conversational intake that interviews a broker or insured, probes the ambiguous exposures that define a hard-to-place risk, and hands the underwriter a complete file instead of a partial one. In a market where E&S premium has grown double digits for seven straight years and submissions only get more complex, the carrier that masters conversational underwriting intake wins on cycle time without surrendering judgment.

That conversational layer is what Perspective AI builds. If you underwrite complex risk and want to see how an AI interviewer turns a messy submission into a structured, audit-ready file, start a new research project, explore how it works for CX and intake teams, or review the studies behind the approach.