The future of insurance underwriting: how AI amplifies judgment without replacing it

Some industry forecasts suggest AI could handle a majority of underwriting tasks within years. But Andrew McMellin, who has spent 38 years in insurance as a broker, underwriter, and MGA operator across London and the US, sees a more nuanced reality. As President of Markel International, he leads a team that has quadrupled its international portfolio analytics team in size in four years. His view is clear: the path forward is not about choosing between human judgment and technology but building a powerful partnership between the two.

That partnership is already producing measurable results, from 300% productivity gains to approximately $30 million in new business from previously inaccessible segments. But it demands a fundamentally different way of thinking about what underwriting tools are for.

Key takeaways

• Data scarcity in specialty markets makes human judgment structurally essential

• Harvey AI in W&I transactional liability generated $30 million in new business from previously inaccessible segments

• Markel Prime sent 8,000 quotes to 500 new brokers in six months using AI-powered proactive outreach

• Portfolio underwriting is shifting to holistic portfolio thinking, supported by analytics teams quadrupled in size

• The intangible risk gap represents the next major growth opportunity for specialty insurers

What does the future of insurance underwriting look like?

The transformation underway is not a wholesale replacement of intuition with algorithms. It is a shift from purely intuition-based decisions to data-enhanced judgment.

In personal lines like motor insurance, huge data volumes spanning decades can drive insights on their own. But specialty markets exist precisely because they sit at the other end of that spectrum. Data is inherently scarce, risks are complex and heterogeneous, and the nuances do not surface in datasets. In specialty and commercial insurance, the human underwriter remains the decision-maker, now armed with better tools.

Key challenges driving underwriting transformation

Data scarcity and the polarization it creates

The fundamental challenge in specialty underwriting is structural. A specialty underwriter working on a complex liability placement may have a fraction of the data available in personal lines, with each risk carrying unique characteristics resisting standardization.

This scarcity has created an unproductive tension between those who believe data should drive all decisions and those who see technology as a threat to the underwriting craft. As McKinsey has noted, AI in insurance works best when it augments human decision-making rather than attempting to automate it entirely. The organizations pulling ahead are those that reject the binary and invest in both.

Meeting evolving client needs

Clients face increasing uncertainty, and a growing share of their risks are intangible in nature. The insurance market has historically excelled at covering tangible risks, but the intangible risk frontier is where the next wave of product innovation will come from. Addressing it requires multi-disciplinary judgment that no algorithm currently provides.

How AI and data analytics are reshaping specialty underwriting

In a recent episode of The Underwriting Intelligence Podcast, McMellin shared specific examples of how Markel is deploying AI across its operations, not as a replacement for underwriting craft, but as an accelerant.

Data extraction and processing

At the most foundational level, AI excels at scraping and extracting data into underwriting systems. Markel applies this in its terrorism business for value extraction and in processing electronic slips into underwriting platforms, freeing underwriters from manual data handling.

Document analysis and information retrieval

The Harvey AI implementation in Markel's W&I (warranty and indemnity) transactional liability book stands out as a breakthrough use case. The tool searches through massive documents and, through carefully crafted prompts, distills the information underwriters need to make decisions.

As McMellin explained: "We can utilize it to go and search through huge documents of data. And by creating the right prompts, the underwriters can distill that information."

The results speak for themselves: a 300% productivity increase and approximately $30 million in new business from market segments that were previously too resource-intensive to underwrite efficiently. The AI did not make the underwriting decisions. It supplied the underwriters with data in a form they could act on.

Proactive market development

The Markel Prime initiative represents a fundamentally different use of AI: proactive rather than reactive outreach. Launched in the European SME market, the program uses AI to scrape publicly available data from the internet, harness it with Markel's pricing tools, and generate quotes for risks like miscellaneous PI coverage.

In six months, the program sent 8,000 quotes to 500 new brokers, reaching clients who had never had access to Markel before. The pricing models behind those quotes are still built and governed by underwriters. AI handles the data gathering and distribution at a scale that manual processes could never match.

Enterprise-wide AI enablement

Markel deployed Microsoft 365 Copilot across its entire company, enabling underwriters, claims teams, and support functions to create their own prompts and workflow improvements.

The evolving role of the specialty insurance underwriter

From individual risk to portfolio thinking

Portfolio underwriting has matured from an emerging concept to a core competency. The approach requires stepping back from individual risk decisions to evaluate how each risk adds to, or detracts from, the broader portfolio. It is about understanding the dynamics of the whole book rather than optimizing each placement in isolation.

Building multi-disciplinary collaboration

Markel's investment tells the story. The firm quadrupled its international portfolio analytics team in four years, building a group that includes portfolio managers, data analytics specialists, and pricing model builders. All exist to serve underwriters, helping them determine how to manage their portfolios over time. This is not a technology team operating in isolation. It is a multi-disciplinary function embedded alongside underwriting, designed to add tools to the underwriter's armory.

Embracing models as tools, not masters

The critical mindset shift for underwriters is treating models as inputs rather than answers. Not every risk fits neatly into a pricing model, and forcing the fit produces misleading outputs. McMellin put it directly: "The models are only ever there as a tool. The underwriter is making a judgment, not the actual pricing model."

With limited data points, models can be made to justify almost any conclusion. CAT models provide a useful parallel: primary modeling firms revise their outputs constantly as new data becomes available, openly acknowledging embedded assumptions. Underwriters should adopt the same mindset, challenging assumptions rather than blindly following outputs that the next version may contradict.

Why specialty underwriting still requires human judgment

Testing intuition against data

The most powerful application of data in specialty markets is not replacing judgment but testing it. An underwriter's intuition might point one way, but historical data could reveal that those assumptions have not held over time. The combination yields better decisions than either approach in isolation.

The subscription market advantage

Whether through surplus lines and admitted markets in the US or the subscription model in London, the market structure enables risk-sharing across multiple participants. The Lloyd's subscription market is particularly well suited for tackling new and intangible risks, bringing together multi-disciplinary specialists to break down complex risks into insurable components. Cyber insurance is the proof point: twenty years ago, many said it could not be insured, and today it is a massive, growing market.

Preparing your organization for the future of underwriting

Invest in analytics capabilities

Build teams with diverse skills spanning portfolio management, data analytics, and pricing model development. Ensure these teams serve underwriters rather than operating as a separate function.

Adopt the right framing for technology

Frame data and AI as additional tools at the hand of underwriters to help them make better judgments. The conversation should always be data enhancing judgment, never data versus judgment.

Create feedback loops

Regularly review model assumptions with underwriting teams. Enable underwriters to tweak models by incorporating their own experience and intuition alongside the latest data.

Focus on emerging opportunities

The intangible risk gap represents enormous opportunity. Insurance penetration remains low even in mature markets, and clients' risk profiles are shifting toward assets the market has historically underserved. Use cyber as the template: what seems uninsurable today could become a massive market within a decade.

Conclusion

The future of insurance underwriting is not a story of humans versus machines. It is a story of humans with machines, where AI handles data extraction, document analysis, and proactive market outreach while underwriters retain the judgment that specialty markets demand. As McKinsey research confirms, the most successful insurers are those combining AI capability with human expertise. Markel's results across Harvey AI, Markel Prime, and enterprise Copilot deployment demonstrate this formula in practice. The winning formula combines technological capability with deep domain expertise.

For underwriting leaders navigating this transition, the priority is clear: invest in analytics capabilities, frame technology as an amplifier of judgment, build feedback loops between underwriters and models, and pursue the emerging opportunities in intangible risk that clients increasingly need covered.

For deeper insights into how leading insurers are balancing AI with underwriting expertise, watch the full conversation with Andrew McMellin on The Underwriting Intelligence Podcast.

Explore how hyperexponential helps underwriting leaders combine data-driven pricing and portfolio analytics with expert judgment to navigate the AI-augmented future of specialty insurance.

Frequently asked questions

Will AI replace insurance underwriters?

No. In specialty and commercial insurance markets, data scarcity is structural, making human judgment indispensable for evaluating complex, heterogeneous risks. AI works best as an augmentation tool, supplying underwriters with data and analysis while they retain the final decision.

How is AI currently used in insurance underwriting?

Insurers like Markel are deploying AI for data extraction and processing, document analysis using tools like Harvey AI, proactive market development through programs like Markel Prime, and enterprise-wide productivity via Microsoft 365 Copilot. In each case, AI supports underwriter decision-making rather than replacing it.

What is portfolio underwriting?

Portfolio underwriting shifts focus from individual risk assessment to evaluating how each risk contributes to the broader portfolio's performance. It demands understanding how line sizes, business mix, and commercial relationships interact to drive volatility and profitability at the portfolio level, enabling underwriters to make individual decisions with the whole book in mind.

What is the intangible risk gap in insurance?

The intangible risk gap refers to the growing mismatch between clients' evolving risk profiles and the coverage available in the market. As businesses increasingly depend on intangible assets, insurers must develop new products to address these exposures, much as the market developed cyber insurance over the past two decades. The subscription market structure, particularly at Lloyd's, is well positioned to tackle this frontier.