Nobody Is Coming to Fix the Mid-Back Office

Blueprint Two promised £800m and delivered a pause. Four supervisors did not pause with it.

The market frame

UK insurance writes $187bn of gross written premium, up 17% from 2022, contributing £61bn to UK GDP and 37% of City GDP (FCA, Regulatory Priorities: Insurance, 2026).

Lloyd's wrote £57.9bn in 2025 with £10.6bn of profit, an 87.6% combined ratio and £49.8bn of capital.

Ireland's non-life market writes roughly €23.3bn. Net of intra-market cession and the Lloyd's-to- company double count, the working addressable UKI P&C pool is approximately £150bn. (calculated estimates, not published.

Three exhibits to the market frame:

• The PRA's 2026 supervision letter records a continuing softening underwriting cycle, pronounced in London wholesale
• Oxbow Partners found the London Market expense ratio fell from 38.4% in 2018 to 35.0% in 2023 while the administrative component stayed flat at 10.2% against 9.8%,
• and rate-adjusted the admin ratio rose to nearly 12%, with almost 60% of syndicates holding at-risk expense bases. while Lloyd's 2026-30 strategy has withdrawn the centre from delivery.

In effect, Rate that was doing the work is stopping. The cost base never improved, the centre stopped, and the liability stayed!

Blueprint Two was launched in November 2020 targeting £800m of savings and full adoption by Q2 2024. In February 2026 it was reported shelved, the market engagement team disbanded.

Phase 1 was rebuilding POSH and LIDS, infrastructure first built in the 1980s, with forty years of undocumented EDI alterations and native knowledge held in the heads of people, no longer in the market . That in summary frames the market.

The problem statement: One Transaction - One set of facts - Four Supervision layers -Ten Destinations

A Continental carrier has a parallel books problem: local GAAP, IFRS 17, Solvency II and tax, reconciled to one supervisor. A UKI group has that plus a parallel audience problem:

Ten destinations, ten definitions, one transaction. That is not inefficiency. It is the mechanism that has produced two PRA fines in one year.

The net effect

• Without a governed data model each destination is sourced independently, by a different person, from a different extract, at a different time
• The differences are then reconciled by hand; resulting in two PRA fines in a single year

On 10 March 2026 the PRA fined U K Insurance Limited, Direct Line's principal underwriter, £10.6m for miscalculating its Solvency II balance sheet during 2023 and 2024

On 16 July 2026 the PRA fined HDI Global SE £4. 16m for three years of materially inaccurate FSCS Liabilities and Fee Tariff submissions

In effect, the regulator has priced the failure!

The root cause:

• The architecture decision was outsourced to the centre, and the centre stopped: Seven years of deferred vendor selection, deferred integration roadmaps and deferred capital allocation, all built around a capability that never hit the road
• Accounting logic sittings downstream of the event and behind the bureau: In a subscription market one risk generates signings across multiple syndicates, and the accounting reconstruction happens after all of them
• No single multi-basis sub-ledger: UK GAAP under FRS 102 and 103 in Schedule 3 format, year-of-account, IFRS 17 at group, Solvency UK, Irish local and Solvency II, and tax – all reconciled offline
• Reconciliation is an event, not a state: High manual rates are an architecture problem, not a people problem. Nobody would love to reconcile by hand, lines and lines of information!
• Regulatory reporting has no data model, across four authorities: Ten destinations for one fact, each sourced independently

in effect, carriers need to file facts; not spreadsheets; Format is the PRA's XBRL taxonomy. Lloyds validates first, then files onward. The PRA's system checks every tag mechanically at submission. Fail the rule, rejected return, rework!

• AI is being asked to fix a 2010 operating model, without redesign first: Roughly 55% of F&A resources sit in transactional processing!

The most expensive finance labour base in Europe and in lockstep with Point Automations on broken reconciliation that produces Faster Broken Reconciliation!

What it costs today:

Illustrative carrier: A £2.5bn GWP UKI P&C group. A £1.4bn Lloyd's syndicate through an owned managing agency, a £0.7bn UK company market platform, a £0.4bn Irish EEA carrier. Six legal entities. Delegated authority approximately 40% of the book. Net earned premium approximately £1.85bn. Combined ratio 93%. Finance function £32m and 280 FTE, being 1.28% of GWP against an APQC top- quartile frontier of 0.57% of revenue.

• Money that never arrived: £10-13m p.a.
• Cost of doing business: £31-46m p.a.
• Working capital, stuck (in unreconciled balances and Reinsurance): £100-150m one time. Of which £45-70m, the realisable slice is the aged paid recoverable plus paid claims not yet billed to the treaty

The objective is to ensure that every financially significant event is emitted once, in a canonical shape, with full dimensions at the point of occurrence

• Layer 1: The event spine Three inbound streams matter in UKI and they behave differently. a. The core platform stream covers direct and open market business: policy, endorsement, premium, claim, recovery, reserve movement. This is the well-behaved one, and if the carrier is mid-cloud-migration on its policy or claims estate, that programme should be the delivery vehicle b. The bureau stream covers signings, closings, central settlement advice and claims movements arriving through market message flows. These arrive in formats shaped by forty years of accumulated convention. The design decision is to normalise them into the canonical model at the onset. This is principle four in practice. c. The delegated authority stream is the hardest and the most valuable, because it is 40% of

the reference carrier's book and roughly 45% of Lloyd's premium. Premium and claims bordereaux arrive in inconsistent formats on variable cycles from coverholders of widely varying sophistication. The current pattern is to receive them, chase the gaps, key the summary and reconcile to cash later. The target pattern should be to ingest at line level on arrival, apply schema inference and mapping against a learned template per coverholder, validate against the binder terms held as structured data, match to risk and to cash, calculate commission and profit commission from the contract rather than from a spreadsheet, and raise only genuine exceptions. This single capability carries the largest share of the recurring value, and it is the thing Blueprint Two was implicitly expected to fix.

Output of Layer 1: a canonical event stream with risk reference, entity, syndicate and year of account, pure year, reporting year, risk code, Solvency UK and CBI lines of business, Schedule 3 class, distribution channel, country of risk, original and settlement currency, binder reference, and the broker and coverholder chain. Every dimension any downstream consumer will ever ask for, captured once, at source.

• Layer 2: The multi-basis ledger One sub-ledger takes each canonical event and posts it simultaneously to every basis the group reports on. One event, one entry, many measurement outcomes.

For a UKI group the bases are UK GAAP under FRS 102 and 103 in Schedule 3 format, year-of- account for the syndicate including RITC, IFRS 17 where the group reports it, Solvency UK, Irish local GAAP and Solvency II for the EEA entity, and tax. The posting rule library is the asset. It encodes, for each event type and each basis, the recognition point, the measurement, the account, the entity and the elimination treatment. Building it from scratch is where most programmes fail, because the rules are not written down anywhere; they live in the heads of the people running the close and in the exceptions their spreadsheets handle.

• Layer 3: the reporting factory

A certified fact store dimensioned at the finest grain that fits consumer needs. Built off the sub- ledger, a mapping layer that projects those facts into each authority's taxonomy, a validation engine that runs each authority's rules before submission rather than after rejection, and a lineage service that can trace any cell in any filing back to the originating events.

a. Lloyd's CMR family from the same certified base that produces the PRA Solvency UK QRTs and NSTs b. the CBI harmonised QRTs and NSTs in the CBI-extended taxonomy, the syndicate annual accounts in Schedule 3 format c. the IFRS 17 disclosure, the IPT and Irish levy returns, and the FCA Consumer Duty and

In essence, a store that holds the same concept that has different definitions across authorities, in an explicit, documented, versioned mapping rule

Reconciliation between filings becomes structural, so QMA, QMB and QMC cannot disagree, because they are projections of one base

Taxonomy change becomes configuration, so NACE 2.1, or the next PRA or CBI taxonomy version, is a mapping update and a regression test rather than a project

In summary, filing effort falls by roughly half. More importantly, the failure mode that produced £14.8m of fines is eliminated through design.

The agentic layer of the solution:

Nine agents, each with a named human control point. Agents execute; humans attest.

Design principles that underpin the solution:

Built around a single organising idea: certify the fact-data once, then let every authority take the version one wants.

• Book it where it happens. Accounting judgement belongs at the point of the event, not three systems downstream. When a line is signed, a bordereau line arrives, a claim moves or a cession triggers, the entry is determined there, with the risk reference, the year of account, the entity, the currency and the ceding structure attached
• Certify once, file many times. Build a single certified fact, dimensioned at the finest grain any consumer needs, then generate every filing from it by projection rather than by extraction. Risk code, pure year, reporting year, Schedule 3 class, Solvency UK line of business, CBI line of business, distribution channel, country, currency, transaction type. Ensure all dimensions are covered, the cost of not having one dimension included when the CBI asks for distribution channel or when the PRA moves to NACE 2.1 is a manual project.
• Close continuously, confirm periodically. Stop treating the close as construction and start treating it as confirmation. If reconciliation runs continuously and breaks are risk-scored and cleared daily, then period end becomes a checkpoint where someone confirms a state that already exists
• Assume no central rescue. Design every dependency on the bureau, on central settlement and on market infrastructure as a replaceable adapter behind your own canonical model, not as a shape your ledger inherits. Blueprint Two is the lesson: any architecture that assumed the centre would change is now seven years stale
• Redesign the work and reskill the resources and govern the machine as infrastructure. Agents are the reason manual activities disappear. And an agent posting to the ledger is production financial infrastructure, so it needs to carry model documentation, version control, change approval, output monitoring and an identifiable human owner from day one

Supervisory expectation: Irish IAF the attestation is personal, so the evidence chain must be personal grade. Lloyd's stated posture is that AI should "strengthen human decision-making rather than replace it."

Operating model and right-shoring:

The organising rule. Work is placed by accountability class, not by price per transaction, rate card. Every activity is classified against a single test: does performing it create, or discharge, a personal regulatory accountability?

Anything that does stays where the accountable individual sits. Anything that does not is placed on capability and cost. This test is what separates right-shoring from the 2010 lift-and-shift, which classified work by whether it could be documented into a procedure.

The five tiers.

Onshore judgement and attestation, London and Dublin. SMF and PCF holders, Chief Accounting Officer, technical accounting policy, RITC and year-of-account judgement, Solvency UK and Solvency II interpretation, the Lloyd's, PRA, CBI and audit relationships, and final sign-off on every return. This population shrinks hard and rises in grade.

Onshore controllership, UK regional. Entity controllers, control testing, Consumer Duty and product value evidence, statutory accounts preparation. Glasgow, Leeds, Bristol, Norwich, Ipswich or Belfast run at roughly 61% of London fully loaded cost with identical regulatory proximity, no data residency question, no visa friction and access to the same qualified population. Most London market carriers have not used this tier at all, which is the cheapest available saving in the whole design.

Nearshore regulatory-proximate, Poland, Portugal or Lithuania. Solvency UK and CBI NST production, XBRL tagging and validation, QMA, QMB and QMC assembly, PMDR and GQD production, IFRS 17 disclosure preparation. EU time zone, EEA data residency for the Irish entity, and a deep Big Four alumni pool in Kraków, Warsaw, Lisbon and Porto. This tier exists specifically because the Irish entity's data should not leave the EEA without a transfer assessment, and because the CBI filing calendar runs on CET.

Offshore capability centre, India. Bordereaux exception ownership and coverholder data stewardship, cash application exceptions, broker credit control execution, reconciliation review and break clearance, reinsurance recoverable chasing, IPT and levy preparation, management reporting production.

The Cost of doing it right first-time: Based on of a £2.5bn GWP UKI P&C group.

AI Advisory leads, not a platformication. An AI-instrumented diagnostic produces evidence in six to eight weeks where traditional assessment produces interviews in six months.

The argument for Build as against Rent and why this is a Global Capability Centre build:

Build against rent, and why this is a Global Capability Centre build. The work is permanent, the accountability is personal, and continuity is the thing that makes it safe, so it belongs inside a centre the carrier owns.

A capability centre supplies capacity and continuity. What it does not need to buy is the thing the carrier already has - The posting rule library, the multi-basis parallel run, the contract boundary judgements, the intercompany elimination across three valuation bases and the auditor-accepted lineage evidence are things this carrier has done repeatedly, cycle after cycle, in front of its own auditors.

The native tacit knowledge, and more importantly the learnings behind it, are invaluable and cannot be replicated. They were earned across closes no provider attended, they exist nowhere else in the market, and today they sit with a handful of specialists rather than with the institution.

A Global Capability Centre is how the carrier stops renting its own knowledge back from individuals and starts owning it at scale its business models, not just for UKI and also Europe- North America- APAC and establishing cross-geo agility and fluidity.

The quantitative comparison:

• £52k of every £90k seat is SG&A, sales, margin and risk loading; strip it across 220 seats and the carrier keeps £11.4m a year, £57m across the five-year window
• AI and agent engineering right-shored inside the centre costs £2.8m a year against £7.6m bought in, and by year five the estate is retuned by 35 people the carrier employs rather than repriced at each renewal
• Vendor management, assurance, procurement and security oversight consume 8-12 carrier FTE and £1.4-2.0m annually, indefinitely
• Five-year cost of ownership £128-144m owned against £172-190m bought, a £42-50m delta, break-even between the two routes at month 26-30