The Lloyd’s Workforce Crisis: How AI Can Preserve Decades of Underwriting Expertise Before It Retires

Lloyd’s of London reported £10.6 billion in profit before tax in 2025. The market is performing exceptionally well. But beneath the headline numbers lies a structural vulnerability that few are addressing with sufficient urgency: the people who built that performance are leaving.

The Numbers

The insurance industry’s demographic challenge is well-documented globally — 400,000 roles at risk of going unfilled, half the workforce over 55 — but the London Market faces a particularly acute version of this problem. Lloyd’s underwriters and senior operations professionals possess highly specialized knowledge that takes 10–15 years to develop: understanding of complex risk classes, familiarity with broker relationships and preferences, institutional memory of how specific accounts have been handled historically, and the judgment to spot anomalies in submissions that less experienced staff would miss.

This knowledge is not written down. It lives in people’s heads. And those people are retiring.

What Gets Lost

When a senior property underwriter at a Lloyd’s syndicate retires, the organization doesn’t just lose a person. It loses:

• Knowledge of how 200 different brokers format their submissions
• Understanding of which data points matter most for specific risk classes
• Pattern recognition developed over thousands of reviewed submissions
• Relationships and context that inform risk appetite decisions
• The ability to spot a problematic submission in seconds rather than hours

Hiring a replacement fills the headcount. It does not replace the knowledge. A new underwriter will take years to develop the same pattern recognition — years during which the organization operates at reduced efficiency and elevated risk.

The Traditional Response

Most Lloyd’s managing agents respond to this challenge with documentation initiatives, mentoring programmes, and knowledge management systems. These are well-intentioned but fundamentally limited. You cannot document intuition. You cannot write a procedure manual for “this SOV looks wrong based on 15 years of experience reviewing similar accounts.”

The knowledge that matters most — the tacit, experiential judgment that separates a competent underwriter from an exceptional one — resists codification through traditional means.

A Different Approach: Learning Through Work

What if the system your underwriters use every day was quietly capturing their expertise as they worked? Not through documentation projects or knowledge-sharing sessions, but through the natural act of reviewing submissions, correcting extracted data, and making decisions.

This is the principle behind Learning Insurance Operations Platforms. Every time an underwriter corrects an extraction — changing a construction type from “Frame” to “Joisted Masonry” because they recognize the building from prior submissions — that correction becomes institutional knowledge. The system learns that for this type of property, in this postcode, from this broker, the construction classification follows a specific pattern.

Multiply that by thousands of corrections across hundreds of submissions, and the platform develops a form of institutional memory that persists regardless of staff turnover. The senior underwriter’s expertise doesn’t retire with them. It lives on in the system, improving extraction accuracy and flagging anomalies for the next generation of underwriters.

Cazimir’s Role

Cazimir is built on this principle. The platform doesn’t just extract data from submissions — it learns from every human interaction. When a Lloyd’s underwriter reviews a complex specialty submission and makes corrections, those corrections train the system for future submissions of the same type, from the same broker, in the same class of business.

The result is a platform that gets measurably better over time — and that preserves institutional expertise as a permanent organizational asset rather than a depreciating human resource.

For Lloyd’s managing agents planning for the next decade, the question isn’t whether to invest in AI. It’s whether to invest in AI that learns — capturing the expertise of today’s workforce before it’s gone — or AI that merely automates, leaving the knowledge gap unaddressed.