IReF: Why the ECB’s 2031 Timeline Still Means Starting Your Data Model Now

The ECB has confirmed that the first official IReF reporting is not until the second quarter of 2031, with a pilot the year before. That distance is being read inside banks as permission to wait. It is the opposite.

In June 2026 the ECB published the main milestones for rolling out the Integrated Reporting Framework. A public consultation on the draft IReF Regulation is expected in the second half of 2027, a pilot phase runs through 2030, and the first live reporting begins in the second quarter of 2031, followed by a parallel period during which the existing collections keep running alongside the new one. Read quickly, that timeline says you have five years. Read properly, it says the one piece of work that determines whether IReF is cheap or ruinous — the mapping from your internal data to a single ECB data model — has to be well underway long before the pilot, and most banks have not started it.

What IReF actually changes

Today a euro-area bank satisfies several ECB statistical collections separately: balance sheet items (BSI), interest rate statistics (MIR), granular credit data (AnaCredit) and securities holdings (SHS). Each has its own regulation, its own templates, its own quirks of definition, and — in most banks — its own reporting pipeline, built at a different time by a different team from a different extract of the source systems. That fragmentation is the status quo IReF is designed to end.

IReF replaces those separate obligations with a single, directly applicable reporting scheme built on one data model. Instead of producing four reports from four pipelines, you submit granular data once, and the required aggregates are derived from it. The promise is a genuine reduction in duplicated effort. The condition attached to that promise is that your data has to be expressed in the terms of the ECB’s model — consistently, with the same instrument meaning the same thing whether it turns up in a balance-sheet figure, a credit record or a rate calculation. That consistency is exactly what fragmented pipelines never had to enforce, because each report was reconciled to itself and to nothing else.

The work is semantic, and semantic work is slow

The mistake behind the “we have time” reading is treating IReF as a reporting-system replacement — a project you scope, procure and deliver in the eighteen months before a deadline, the way firms treat a new template. It is not that. The expensive part of IReF is not generating the output; vendors will sell you engines that do that. The expensive part is establishing, for every attribute the model requires, where in your estate the authoritative value lives, what it means, how it is currently transformed on the way to each existing report, and why those transformations disagree with one another.

That is a data-lineage and semantic-mapping exercise, and it moves at the speed of institutional knowledge, not the speed of a delivery plan. It surfaces the counterparty that is classified one way for AnaCredit and another way for BSI. It surfaces the product whose maturity is derived differently in two systems that both claim to be the source. It surfaces the fields that no living person can fully explain because the person who built the mapping left in 2017. None of that is solved by buying software. It is solved by people reconstructing meaning across the estate, and there is a hard limit on how fast that can be done, because the people who hold the knowledge also have day jobs keeping the current reports out the door.

This is the same trap I described for the EHDS certification timeline: a date years out that looks like breathing room but is actually a data-model problem whose long-lead work has to start immediately. IReF is that pattern in the reporting domain. The ECB has even provided a head start in the Banks’ Integrated Reporting Dictionary (BIRD), a voluntary, shared description of how to derive the required output from banks’ internal data. BIRD is useful precisely because it assumes you have first done the hard part — knowing your own inputs well enough to map them.

The internal case for deferral, and how to counter it

The argument for waiting always sounds responsible. The Regulation is not adopted; consulting on it opens in 2027; specifics may change; why build against a moving target. Each point is true and none of them justifies inaction, because the work you should start now does not depend on the final template at all. You do not need the adopted text to inventory where your balance-sheet, credit, rate and securities data originates, to document how each current report transforms it, or to find the places where two pipelines disagree about the same fact. That reconciliation has to happen whatever the final model looks like. It is the foundation, and it is entirely within your control today.

The honest counter to the deferral case is this: the consultation and the pilot are checkpoints that assume the foundation exists. A firm that arrives at the 2030 pilot without having mapped its estate is not testing IReF — it is discovering, under a deadline, how inconsistent its own data has always been. That discovery is the most expensive way to learn it, because you are now paying reconciliation costs and pilot pressure at the same time, with regulatory visibility on the result.

What to begin now

• A source-of-truth register for the four domains. For balance sheet, credit, interest rates and securities holdings, name the authoritative system for each attribute — and resolve the cases where more than one system claims it.
• Documented lineage from source to each current return. Every transformation between a source field and a submitted figure, written down. Where you cannot reconstruct it, that is a finding, not a gap to paper over.
• A reconciliation of cross-report inconsistencies. The definitional conflicts between AnaCredit, BSI, MIR and SHS are your IReF backlog. Enumerate them before the model forces you to.
• A granular-first data foundation. IReF derives aggregates from granular records. An estate that can only produce pre-aggregated returns has architectural work to do, and open table formats are part of that data-strategy decision.

None of this is IReF-specific in the narrow sense, which is the point. It is the data governance you should be able to demonstrate anyway, brought forward by a regulation that will eventually inspect it. The banks that treat 2031 as a start date will spend 2030 in reconciliation triage. The ones that treat 2026 as the start date will arrive at the pilot testing a framework, not excavating one.

The deadline is not the moment the work is due. It is the moment the work becomes visible. Those are years apart, and the gap is where the cost is decided.