Where Should London Market Carriers Start After Lloyd’s Blueprint Two?
Blueprint Two promised to fix the market’s data before it reached you. Agentic AI lets carriers work with that data as it actually arrives through the Electronic Claims File (ECF), and write back a better record than the one they received.
11 min read
For six years, a common answer to “What’s your London market technology strategy?” was some version of “We’re waiting for Blueprint Two.” On 19 March 2026, that answer stopped working.
The question carriers should ask now is “What does our claims operation do with what comes through ECF, and what does it send back?” The answer to that question used to require a market-wide programme. It no longer does.
What you will learn:
- What Lloyd’s sunset in March 2026, and which parts of the infrastructure continue
- The bet Blueprint Two made about structured data, and why agentic AI reverses it
- Why most of what matters on a London market claim arrives unstructured
- How retrieving, reading and ECF Write Back turn an ECF connection into a system of action
- One reserve advice, handled end to end
- Where to start, in what order, without waiting for the market
Key definitions
These are the terms that shape London market claims after Blueprint Two. Each one is written to stand on its own.
Term | Definition |
|---|---|
Blueprint Two | Lloyd’s market-wide digital transformation programme, launched in 2020. Its name and original vision were sunset in March 2026. |
CLASS | The Claims Loss Advice and Settlement System, where brokers submit claims and transactions as structured data. |
IMR | The Insurers’ Market Repository, the central store for placement and claims documents, organised by UMR (policy) and UCR (claim). |
ECF | The Electronic Claims File, the interface Lloyd’s and IUA carriers use to review and respond to claims, combining CLASS data with IMR documents. |
Claims agreement party | The insurer authorised to agree a claim for the market under the Claims Lead Arrangements. |
Claims workflow triggers | Structured ECF outputs that signal claim events (new claim, update, completion) so carrier systems can consume them automatically. |
ECF Write Back | Sending a carrier’s approved response, data updates and rationale back to the market record.The market capability that lets a carrier send its approved response, data updates and rationale from its own claims system back to the central claims record. |
System of action | A claims system that uses the record to move the claim forward, not only to store it. |
What happened to Lloyd’s Blueprint Two, and what’s still standing?
Lloyd’s sunset Blueprint Two’s single-platform vision on 19 March 2026. CLASS, IMR and ECF keep running, cloud re-platforming is expected by 2028, and Lloyd’s keeps setting data standards. What happens to a claim message once it reaches a carrier is now that carrier’s decision.
Blueprint Two launched in 2020. Its phase one cutover slipped from 2024 to 2025, then to 2026, and full re-platforming is now expected by 2028.
Chief Executive Patrick Tiernan was direct about why: “The future of the Lloyd’s market infrastructure is not one platform, mandated across hundreds of market participants,” he said, as reported by Reinsurance News. Lloyd’s now describes the goal as “an ecosystem of intelligent solutions, built on common data standards and interoperability.”
Lloyd’s keeps the job of setting standards and organising market data, working with Velonetic, DXC and the IUA on the shared pipes. CLASS, IMR and ECF keep running. Everything that happens after a claim message reaches your desk belongs to you.
Brokers have already moved. In Guidewire’s London Market Tech Barometer 2026, 78% of brokers said they were pursuing their own technology strategies regardless of the market timeline, and 78% called insurer technology decisive or highly significant when choosing where to place risk. For carriers, claims technology is now part of the commercial relationship.
The bet Blueprint Two made, and why agentic AI reverses it
Blueprint Two assumed the market would structure the data first and carriers would automate on top. Agentic AI reverses that order: it reads claim documents and free text as they arrive through ECF and creates the structure inside the carrier’s own claims operation.
The logic of Blueprint Two was sound for 2020. Get every participant onto common data, structured at the point of entry, and automation becomes possible downstream. Structure first, automate later.
That order made sense when software could only act on fields. A rules engine can compare two reserve figures. It cannot read an adjuster’s report and tell you the figure is wrong.
Large language models changed that constraint. Software can now read a report, a broker narrative or a coverage letter, extract what matters, and say where each fact came from. Structure no longer has to arrive with the data. It can be created on arrival, inside your own claims operation.
That changes what a carrier’s ECF integration is for. Under the old model, the integration was plumbing: it moved messages into a queue for people to read. Under the new one, it is the point where the claim is understood, and the point from which the market record is kept complete.
Blueprint Two promised to fix the data before it reached you. Agentic AI lets you work with the data as it actually arrives, and write back a better record than the one you received.
The industry reaction pointed the same way. Ben Rose, co-founder and president of Supercede, told Insurance Times:
“The lesson from Blueprint Two is that the industry doesn’t need another moonshot. Incremental improvement across the market will ultimately move the dial far more than one enormous programme ever could.”
Ben Rose, Co-founder and President, Supercede
What actually comes through ECF on a London market claim?
ECF combines structured CLASS data (reserves, payments, dates and references) with IMR documents and free text. The structured part is thin: the material that drives decisions sits in adjuster reports, broker narratives, coverage correspondence and market comments, and today a person reads all of it.
CLASS carries the claim as data: reserves, amounts paid, dates, references. That is what most integrations consume, and it is the part Blueprint Two set out to improve.
The substance sits elsewhere. It is in the IMR documents linked to the UCR and in the free text around each movement.
What arrives | Form | What it tells you |
|---|---|---|
CLASS movement | Structured fields | That the advised reserve or payment changed, and by how much |
Broker narrative | Free text | Why the broker says it changed |
Loss adjuster report | Long document | What the loss is likely to cost, often in more detail than the advised figure |
Coverage correspondence and counsel’s opinion | Documents | Whether the loss is covered, and on what reading of the wording |
Market comments | Free text | What the leader decided and why |
Bordereaux and TPA reports | Spreadsheets and documents | How delegated claims are being handled |
Reading that material is where the cost and the delay sit. A handler opens the advice, opens the documents, compares them with the existing file, decides, keys the response, and writes the comment. A new connection to the central services does not remove any of those steps. If the integration only moves messages faster, it moves the same manual work faster.
This is the gap an agentic AI closes: it does the reading and prepares everything around the decision, and the decision stays with a claims professional.
Retrieve and read: how agentic AI turns ECF traffic into your own view of the claim
Agentic AI should retrieve, read and classify every incoming ECF message, compare it with your own view of the claim, and route it by authority and evidence before a claims professional opens it.
The read side of the integration does four things for each message:
- Classifies it. New claim, reserve movement, settlement request, query, document-only update. ECF claims workflow triggers tell your system an event happened; classification tells it what kind of work it is.
- Extracts what matters from the documents. Figures, dates, parties, causes, coverage points, recommendations. Each extracted fact links back to the page it came from.
- Compares it with what you already know. The advised reserve against the adjuster’s estimate. The new narrative against the last one. The loss against the policy wording and your existing assessment.
- Routes it. In-authority and well documented, with no conflict in the evidence: prepared for straight-through handling. Anything else: to the right person, with the file already summarised and the conflict highlighted.
This is where a system of record becomes a system of action. A system of record stores what happened. A system of action uses that information to move the claim forward, and records each step as it goes. Read more about the shift from a system of record to a system of action in our “The Agentic Era of Claims” review article.
The practical effect is that a senior handler stops being the first reader of every message. They become the decision-maker on the messages that need one.
ECF Write Back: the market record as a by-product of claims handling
ECF Write Back lets a carrier send its response, CLASS updates and decision rationale back to the market from its own claims system. When agentic AI drafts all three and a claims professional approves them, the record Lloyd’s expects is produced as part of handling the claim.
Retrieving and reading claim files alone saves time. ECF Write Back is what connects the agentic AI to the obligations Lloyd’s places on the claim.
London market claims run on a shared electronic record (ECF). Lloyd’s says no more than two lead insurers are needed to agree a claim, however many syndicates sit on the policy, and in more than 85% of cases one leader agrees on behalf of the market.
That structure is why the record matters. Lloyd’s guidance on the Lloyd’s Claims Scheme expects the leading managing agent to keep central claim information complete and accurate for followers, and to record the rationale for decisions in the ECF comments. If you lead, your claims record is the market’s claims record.
In practice, that rationale is often the thinnest part of the file. The real reasoning sits in an email thread, a reserving committee note or a handler’s head.
Write Back closes that gap. Once the read side has done its work, the same system drafts everything that goes back to the market:
- The response to the broker: agree, query or decline.
- The CLASS field updates that go with it.
- The ECF comment, stating the decision, the evidence and the reasoning, with references to the documents relied on.
A claims professional with the right authority reviews and approves. The approved response, the data and the rationale go back together.
The result is a market record that is more complete than most are today, produced as part of handling the claim rather than as a separate job. The evidence your board, your followers and Lloyd’s Market Oversight ask for (why the reserve sits where it does, who agreed the claim and on what basis) already exists, in the place they would look for it.
Teams that rebuild the record afterwards from emails and spreadsheets do the work twice: once to handle the claim, then again to prove how it was handled.
How agentic AI handles one claim message, end to end
Here is one reserve advice on a claim you lead: agentic AI does the reading and drafting, and one claims professional agrees.
A broker advises a reserve increase from $5 million to $7 million, with a narrative and an updated loss adjuster’s report in IMR. Today, a senior handler reads all of it, compares it with the file, decides, keys the response and writes the comment.
The agentic AI reads the report, finds that the adjuster’s estimate points to $10 million, and marks the advice as a conflict that needs a person, not a straight-through item. It drafts a query to the broker, the CLASS update and an ECF comment that cites the relevant pages. The handler reviews a prepared file, edits if needed and approves. One human decision, and the market record carries the reasoning.
An advice with no conflict and within authority follows the same path, but under the straight-through rules you have written down for that message type.
Bring Agentic AI Into London Market Claims.
See how Five Sigma’s AI-native CMS helps move claims forward while keeping claims professionals in control across lead, follow, and oversight workflows.
Design around your role on each claim: lead, follow or delegated
Leaders gain most from ECF Write Back, followers from reading every movement on their line, and delegated oversight from reading bordereaux and TPA reports at volume. Your design should follow your role, claim by claim.
Your role | Where the value is | What the system should do |
|---|---|---|
Lead | ECF Write Back | Draft the agreement, CLASS updates and ECF rationale for approval, so followers see a complete record without extra work |
Follow | Read | Read every movement on your line, compare it with the evidence, and flag where the evidence points to a different outcome |
Delegated (coverholder or TPA) | Read, at volume | Check bordereaux and TPA reports against the agreed data and service standards, and surface exceptions for oversight |
When you follow, you are generally bound by a decision someone else agreed to. That makes the read side essential. If a broker advises a move from $5 million to $7 million and the adjuster’s report points to $10 million, you want to know today. You can then hold an additional case reserve or adjust your IBNR view, and raise it with the leader, rather than finding out at quarter-end.
When a TPA or coverholder handles the claim, your job is oversight. Lloyd’s 2026 oversight priorities put delegated claims handling and data quality near the top. Reading what your delegates send, at the volume they send it, is the only way to oversee it continuously rather than by sample.
Four guardrails for agentic AI in London market claims
Agentic AI that writes to the market record needs four guardrails: people agree claims, every extracted fact is traceable, decision rights are written down per message type, and the audit trail is captured as the work happens.
Agentic AI that writes to the market record has to be designed for accountability from the start. These are the questions to ask of any approach, including your own.
- People agree claims. AI prepares. Agreement stays with a claims professional acting within their delegated authority. The AI drafts the response, the data and the rationale; a named person approves it. Automated handling is limited to items you have explicitly defined as in-authority and low-risk.
- Every extracted fact is traceable. A figure read from a 40-page adjuster’s report should link to the page and paragraph it came from. If a handler cannot check a number in one click, they should not be asked to rely on it.
- Decision rights are written down per message type. For each kind of message, define in advance who decides, at what threshold and on what evidence. Lower-value, in-authority, well-documented items are candidates for straight-through processing. Coverage disputes, large losses and litigation go to senior professionals with the file prepared.
- The audit trail is captured, not reconstructed. Every material decision, its evidence and its approver are recorded as part of the workflow. That supports Lloyd’s claims management principles on reserving, timely insight, third-party performance and executive oversight.
Lloyd’s rules are technology-neutral. These guardrails make an agentic AI process easier to evidence than a manual one.
Why keep the market connection as a separate layer?
The central services are expected to be re-platformed by 2028, so the interfaces carriers use will change. A separate, standards-based connection layer means those changes don’t force a rebuild of your claims workflow.
The pipes will change again. Your claims intelligence should not have to.
The central services are expected to be re-platformed by 2028, and the interfaces carriers use to retrieve IMR documents and respond through ECF will change with them. If your reading, decisioning and writeback logic is wired directly into today’s interfaces, every change in the pipes becomes a rebuild of your claims operation.
Keep the market connection as its own layer, built on the standards the market continues to maintain, with ACORD messaging at its core. Your claims workflow and agentic AI sit above it and consume a stable, internal view of each claim, linked by UCR.
The same separation helps with operational resilience. Under Lloyd’s Principle 12, you remain responsible for your impact tolerances even when a third party runs part of the service. A clean line between connection and workflow makes that easier to test and to evidence.
What should London market carriers do first? A seven-step plan
Start by mapping what comes through ECF today and who decides on it, then build the market connection to serve that design.
Christopher Croft, Chief Executive of LIIBA, called the moment “an opportunity for radical thinking.” Here is a practical order for it:
- Map your message mix. Pull a period of ECF and CLASS traffic and count it by type, value, line and role. Most carriers have never seen this view of their own claims work.
- Find where the reading happens. For each message type, note which documents a handler has to open before deciding. That is where agentic AI earns its keep first.
- Write down decision rights per message type. Straight-through, monitored or judgement, and at what thresholds. Do it separately for lead, follow and delegated claim operations.
- Design the exception paths before the happy path. Know who gets a file when automation cannot decide, and what that person sees when it arrives.
- Define your writeback standard. Agree what a complete ECF comment contains, so every response the system drafts meets it.
- Build the connection layer to feed that design. On market standards, separate from your workflow.
- Measure touches and time to decision per message. Items closed per handler rewards the old model. Touches per message shows whether the new one is working.
Key takeaways
- Lloyd’s sunset Blueprint Two’s single-platform vision on 19 March 2026. CLASS, IMR and ECF continue, and re-platforming is expected by 2028.
- Blueprint Two bet on structuring data first and automating later. Agentic AI reverses that order: carriers can create structure on arrival, inside their own operation.
- The structured part of a London market claim is thin. The substance is in adjuster reports, broker narratives, coverage correspondence and market comments.
- Reading turns ECF traffic into your own view of each claim. ECF Write Back turns that view into a complete market record, approved by a person with authority.
- Leaders gain most from ECF Write Back, followers from reading, and delegated oversight from reading at volume.
- Start with your own message mix and decision rights, then build a separate connection layer to serve them.
See Five Sigma’s AI-native CMS in action.
Five Sigma’s AI-native CMS brings AI automation into the claims workflow built for London market carriers.
Frequently asked questions
What happened to Lloyd’s Blueprint Two?
Lloyd’s sunset Blueprint Two’s name and original vision on 19 March 2026 after a review. Cloud re-platforming continues, now expected by 2028, with Lloyd’s focused on data standards and interoperability.
Does ECF still work after Blueprint Two was sunset?
Yes. CLASS, IMR and ECF keep operating, and claim messages keep flowing between brokers and carriers as they do today.
How can AI help with London market claims?
Agentic AI can classify incoming ECF messages, read IMR documents such as adjuster reports, compare advised reserves with the evidence, and draft the response and ECF comment for a claims professional to approve.
Can AI agree a London market claim?
No. Agreement stays with a claims professional acting within delegated authority. AI prepares the file, the response and the rationale; a named person approves them.
What is ECF Write Back?
ECF Write Back is the market capability that lets a carrier send its approved response, CLASS updates and decision rationale from its own claims system back to the central claims record, so followers and reviewers see what was decided and why.
How should a following market monitor decisions made by the lead?
Read every movement on your line against the evidence. Automation can flag advices where the documents point to a higher outcome, so you can adjust your own reserve view and raise it with the leader early.
Does Lloyd’s require claims to be managed digitally?
Lloyd’s rules are technology-neutral. Most claims already run electronically through ECF, and leaders are expected to keep central claim data complete and accurate for followers, which is hard to evidence manually at scale.
Why should carriers keep one central digital claim record?
Because Lloyd’s expects leaders to keep central claim data complete and accurate for followers, and oversight depends on evidence. One record, synced with ECF, keeps decisions, reasons, and approvals together where followers, auditors, and the board can find them.
Michael Krikheli
Co-Founder & CTO, Five Sigma
Michael Krikheli co-founded Five Sigma to bring AI-native automation to P&C claims past rule-based automation into agentic AI that handles the routine work on every claim, with people engaged where judgment matters.
