Regulatory reporting has long been one of the most expensive and least loved activities in a bank. Hundreds of templates, overlapping definitions and quarterly deadlines have produced large teams, long chains of spreadsheets and a steady flow of resubmissions. For the first time in a decade, the regulators themselves are changing the design, and that gives banks a reason to rebuild rather than patch.
The regulators are redesigning the pipes
In the UK, the Bank of England and the FCA have run their Transforming Data Collection programme since 2021. The Bank has said that simplifying and consolidating collections could save the financial sector up to £2 billion a year. Its cloud platform processes more than 30,000 statistical series from more than 300 institutions3. The first concrete cut for banks came in 2025. The PRA proposed deleting 37 reporting templates, which it estimated would save banks £26 million a year1, and confirmed the deletion in December 2025 with effect from 31 December. The 37 comprised 34 FINREP templates, two COREP templates and one PRA-specific return2.
The first cut is concentrated in financial reporting
Banking reporting templates deleted by the PRA from 31 December 2025, by framework (templates)
Note: PS27/25, first deliverable of the PRA's Future Banking Data programme; the PRA estimated annual savings of £26 million.
In the euro area, the ambition is larger and the timeline longer. On 8 June 2026 the ECB set out the milestones for its Integrated Reporting Framework (IReF), which will consolidate banks' statistical reporting into a single standardised framework. A public consultation on the draft regulation is planned for the second half of 2027, a one-year pilot from the second quarter of 2030, and official reporting from the second quarter of 2031, with a year of parallel running4. Alongside it, the Banks' Integrated Reporting Dictionary (BIRD) has published full data lineage for FINREP and asset encumbrance, is adding AnaCredit transformation rules and is analysing COREP. It now publishes its content under an open-source licence5. The EBA and ECB's Joint Bank Reporting Committee has made semantic integration, meaning common definitions across statistical, supervisory and resolution reporting, its 2026 priority6.
Europe's reporting overhaul runs on a long clock
Selected milestones for UK and euro-area bank reporting reform
| Date | Milestone |
|---|---|
| Jun 2021 | EBA identifies 25 recommendations that could cut institutions' reporting costs by 15–24% |
| 31 Dec 2025 | PRA deletes 37 banking reporting templates |
| 2026 | JBRC focuses on semantic integration and common definitions |
| H2 2027 | Public consultation on the draft IReF Regulation |
| Q2 2030 | One-year IReF pilot reporting phase begins |
| Q2 2031 | Official IReF reporting starts, with one year of parallel reporting |
Note: Also draws on EBA (2021), PRA PS27/25 (2025) and EBA/JBRC (2026). IReF dates are subject to adoption of the IReF Regulation.
The prize is real but should not be overstated. The EBA's 2021 cost-of-compliance study found that its 25 recommendations could cut institutions' reporting costs by 15–24%, including €188–288 million for small and non-complex institutions7. Those savings depend on banks changing their own data supply chains, not only on regulators deleting templates.
Why returns go wrong
The PRA has been explicit about what goes wrong. Its enforcement cases and its 2021 ‘Dear CEO’ letter identify recurring root causes. These are inaccurate and poorly formalised interpretation of rules, relevant data missing from reporting systems, highly manual and complex processes, weak ownership and governance, under-resourced teams and limited independent assurance8. In one case, a UK bank adjusted its risk-weighted assets by about £900 million after applying the wrong risk weight to some commercial loans, and was later fined £5.4 million for weaknesses in its reporting governance and controls9.
Each of these root causes is a data or knowledge problem before it is a technology one. That is why AI, used carefully, fits the problem better than it first appears.
Where AI helps, and where it does not
- Rule interpretation. Models can read regulatory text, instructions and Q&A, and draft a structured interpretation that maps each template cell to its definitions, filters and sources. Experts review and approve it. The output is a documented, versioned interpretation, which is exactly what the PRA found missing8.
- Reconciliation of returns. Agents can reconcile each return to the general ledger, to other returns and to prior periods, investigate breaks and propose explanations with evidence, leaving the reviewer to approve or reject.
- Variance commentary. Generating the first draft of period-on-period explanations, grounded in the underlying movements, removes hours of manual narrative work before sign-off.
- Change impact. When a regulator publishes new or deleted templates, AI can trace which data elements, rules and controls are affected, provided lineage exists.
What AI cannot do is make up for missing lineage. A model asked to explain a number it cannot trace will produce a fluent answer that no one can verify. That is why the regulators' own direction of travel, towards common definitions, open dictionaries such as BIRD and granular data, matters so much for banks' AI plans56. Banks already use AI widely. In the Bank of England and FCA's 2024 survey, 75% of UK financial firms were using it10. Few yet apply it to the reporting chain with the controls that chain requires.
Controls matter as much as capability. Every AI-generated interpretation, reconciliation explanation or commentary draft should be stored with its inputs, the model version and the human who approved it. That turns the reporting process into an evidence trail an internal auditor or supervisor can follow, which is the independent assurance the PRA found lacking in weaker firms8. It also lets the bank show which parts of a return were prepared by a person, which by an agent and who signed off.
Sequencing differs by footprint. UK banks can bank quick wins now, as the PRA removes templates and consults on reporting principles under its Future Banking Data programme12. Euro-area banks have a longer runway to IReF, but the data model work, including mapping to BIRD, takes years and is best started while current returns still provide a baseline for parallel running45.
The timing works in banks' favour. UK reform is under way now, and the euro-area reset will take most of the decade. Banks that invest in lineage and common definitions today will be ready for IReF and able to use AI safely in the meantime. Banks that do not will run two reporting stacks in parallel and pay twice.