How AI Is Rewriting Speciality Insurance in Lloyds of London.

The Lloyd's of London market has survived plague, fire, terrorism, and financial crisis. It may be that the most transformative force it has ever encountered arrived not with a bang, but with a prompt.

For three hundred years, the model has been essentially the same: a broker walks a risk around the Room, an underwriter reads a paper slip, writes a line, and the market moves. The intelligence has always been human. The pace has always been human. And the cost has always reflected that.

That is now changing — faster than most people in the market appreciate.

In April 2025, the Lloyd's Market Association surveyed 81 firms including 45 managing agents. The headline result was stark: only 14% have deployed agentic or generative AI in underwriting. The other 86% are either watching, experimenting without commitment, or actively choosing to wait. What makes this statistic alarming is not the 14%. It is what the 14% are discovering — and how fast the gap between them and the rest of the market is widening.

What Is an AI Agent, and Why Does It Matter for Underwriting?

Most underwriters have encountered AI in its familiar forms: a chatbot that answers policy queries, a model that scores fraud risk, a dashboard that shows exposure data. These are useful tools. They help humans do their existing jobs slightly faster.

An AI agent is categorically different. It does not wait to be asked a question. It receives a goal — ‘process all incoming submissions today, score them against appetite, and draft a referral memo for anything that needs senior review’ — and it executes that goal autonomously across multiple systems, making decisions, handling exceptions, and escalating only what genuinely requires human judgement.

The shift is from AI as a tool to AI as a colleague. A colleague that works at machine speed, never misses a detail, operates 24 hours a day, and costs a fraction of a human hire.

‘We have a chance to consider how the complexities of the London specialty market can be a magnet, rather than a barrier, for deployment of intelligence as a service.’ — Rob Myers, LMA Consultant, April 2025

The Transformation Is Already Happening — In Your Competitors’ Buildings

This is not theoretical. The deployments are live, the metrics are published, and the competitive implications are real.

Hiscox: 99% Reduction in Quote Turnaround

In August 2024, Hiscox went live with an agentic system for their Sabotage & Terrorism line. Built on Google’s Gemini LLM via Vertex AI in partnership with a company called Hailo, the system reads incoming email submissions, extracts 15 or more data points, cleanses and geocodes statement-of-values addresses, and produces a structured risk profile — entirely autonomously.

The result: quote turnaround time went from three days to three minutes. That is a 99% reduction. The same underwriting team is now capable of handling a dramatically higher submission volume without adding headcount.

Ki Insurance: Underwriting Without an Underwriter

Ki, the algorithmic follow syndicate backed by Brit Insurance and Google, took the concept further. Ki’s platform reads live slip data from the Placing Platform Limited (PPL) digital marketplace, checks each risk against a pre-defined appetite model, and autonomously provides follow capacity — without any human underwriter reviewing the individual risk.

It binds in seconds. Capacity providers including Aspen, Beazley, and Travelers have added their lines to the Ki platform. The system now writes business around the clock on risks that match its model.

Apollo: Algorithmic Follow Across Three Lines

Apollo Syndicate has deployed Artificial Labs’ Smart Follow product across Marine Hull, General Aviation, and Marine Cargo. The agent reads CDR-compliant slip data from PPL and autonomously decides whether to follow each risk, within pre-set appetite and authority parameters. A human underwriting director sets the rules. The agent executes them at scale.

65% of Lloyd’s managing agents have NOT yet deployed AI in underwriting or claims (LMA Survey, April 2025) — representing the largest untapped efficiency opportunity in the market.

£57bn+ Lloyd’s GWP in 2024 — the scale of the market that is being transformed by these tools.

The Eight Agent Types Every Underwriter Needs to Understand

The confusion in this market often comes from treating ‘AI’ as a monolithic category. In practice, the agentic era requires a stack of specialised agents, each doing one part of the underwriting workflow extraordinarily well. Here is what the complete picture looks like:

Submission Intake & Triage Agents: These are the front door. They receive broker emails and documents, extract structured risk data, score against your appetite model, and route to the right underwriter — without anyone reading an email. Sixfold, Cytora, and Artificial Labs’ Ava are the leading tools. Zurich North America reports saving up to two hours per submission across 200+ underwriters using Sixfold.

Document Extraction Agents: The MRC slip, the cover note, the broker presentation — these agents read them all and extract every material data point into structured, CDR-compliant formats. Eigen Technologies has been doing this at Hiscox since 2018. V7 Go does it with visual grounding, linking every extracted field back to its exact source location in the document — critical for audit trails.

Pricing and Risk Scoring Agents: These agents take the extracted risk data, enrich it with third-party sources (satellite imagery, loss histories, supply chain data), and produce an indicative pricing range aligned to your book’s historical performance. Hyperexponential and Federato are the leading platforms; several Lloyd’s syndicates use both.

SOV Processing Agents: Statement-of-values cleaning is one of the most painful, time-consuming tasks in property underwriting. Archipelago’s SOV Manager auto-maps columns from any format, geocodes every address, and flags outlier values — in minutes rather than days. Hiscox’s internal Hailo tool does the same for S&T submissions.

Policy Comparison and Wording Agents: When a renewal arrives, someone needs to compare the expiring policy against the new terms, flag every exclusion change, identify new endorsements, and spot coverage gaps. Today this takes hours per policy. Qumis does it with legal-grade semantic reasoning. Chisel AI does it for $79 per user per month. Neither requires IT integration to start.

Portfolio Monitoring Agents: These agents watch the live book continuously, alerting underwriters when a new risk would push accumulations past a threshold, when a class is running ahead of plan, or when the rate level on a line is drifting below adequacy. Federato and Cytora both offer this capability. It replaces the quarterly spreadsheet review with real-time intelligence.

Compliance and Data Quality Agents: DQPro, used by 45% of the Lloyd’s market, automatically checks that bound terms match what sits in the policy administration system — catching data entry errors before they become E&O claims. As Blueprint Two mandates CDR compliance, these data quality agents become table stakes.

Algorithmic Follow and Auto-Bind Agents: The most advanced tier: systems like Ki and Artificial Labs Smart Follow that can autonomously bind risks within pre-set parameters without a human reviewing the individual submission. Not appropriate for every risk or every syndicate, but transformative for follow lines where the underwriting decision is essentially whether a risk fits a pre-defined model.

74% of Lloyd’s firms use AI for data extraction. Only 14% have deployed agents inside the underwriting workflow itself. That gap — between extracting data and acting on it autonomously — is where the competitive advantage is being built.

The Economics: Why This Is an Existential Issue in a Softening Market

For the past four years, rate hardening provided cover for operational inefficiency. When rates are up 7–10% per annum, you can afford to be slow. When rates have softened to +0.3% (as in 2024), the only way to improve the combined ratio is to reduce the expense ratio.

McKinsey’s research shows that AI-embedded insurers deliver Total Shareholder Return six times higher than laggards over a five-year period. In practical terms, the first-mover cohort in this market — Hiscox, Beazley, Ki, Apollo, Lockton, BMS, Marsh — is already operating at structurally lower cost per pound of premium than firms that have not deployed.

That gap compounds. Every submission the Hiscox S&T team processes in three minutes instead of three days frees underwriter capacity for higher-value risks, better client relationships, and more sophisticated portfolio management. The cumulative advantage of 12 to 18 months of this is not easily closed by a competitor that eventually deploys the same tools later.

In a softening market, the combined ratio is the competitive battleground. Firms with agentic AI will operate at 70–75%. Those without face 90%+. In a £57bn market, that gap is existential.

What Should a Managing Agent Do Right Now?

The answer is not to attempt to deploy all eight agent types simultaneously. The evidence from successful deployments suggests a sequenced approach:

• Start with the highest-volume, most repetitive workflow. For most Lloyd’s underwriters, this is submission triage and SOV processing. A single agentic tool addressing either of these will demonstrate ROI within 60 to 90 days — enough to build internal confidence for broader deployment.
• Choose tools that require no IT integration to start. The single largest barrier to adoption cited in the LMA survey (46% of respondents) was integration with legacy systems.
• Apply for Lloyd’s Lab Cohort. The Lab has produced Sixfold, Artificial Labs, and dozens of other tools now live in the market. Participating syndicates get direct access to the most promising agentic tools in a structured evaluation environment, with Lloyd’s market relationships built in.
• Set your appetite rules explicitly before deploying agents. The most common mistake is deploying an agentic tool without having articulated the underwriting rules it should enforce.
• Measure and publish your results internally. The cultural barrier to AI adoption in Lloyd’s is as significant as the technical one. A credible internal case study — ‘We deployed X, it saved Y hours per week, and here is what the underwriting team is doing with that time’ — is the most effective change management tool available.

Conclusion: The Agent Has Already Entered the Room

The Lloyd’s market has always been defined by the quality of its underwriting judgement. That will not change. What will change is the ratio of time underwriters spend on judgement versus administration — and for firms that adopt agentic tools, that ratio will shift dramatically in favour of the former.

The 14% who have deployed are not experimenting. They are building a durable competitive advantage. The 65% who have not yet started are not protecting anything. They are ceding ground every day.

The agent has entered the room. The question for every managing agent in this market is not whether to let it in — it is how quickly they can put it to work.

Sources: LMA AI Survey, April 2025; Lloyd’s Annual Report 2024; McKinsey ‘AI in Insurance: Implications for Investors’, January 2026; Hiscox/Hailo case study; Ki Insurance public documentation; Artificial Labs Series B announcement, January 2026; Sixfold Series B announcement, January 2026; Gallagher Re Global InsurTech Report Q4 2025.