An anonymised quarterly benchmark of ECB on-site inspection findings across Ezelman mandates — by portfolio type, finding severity, and year. The only public data set that tells European banks what their peers are actually being asked in the inspection room.
Industry benchmarks circulate in the 60–90 findings range. On Ezelman-supported inspections, the number is under 40 — because the 80 the bank would otherwise get are pre-empted during readiness.
Observed finding counts per inspection, rolling 4Q average, across Ezelman-supported ECB/SSM on-site inspections in 2024–2026.
| Theme | Ezelman-supported | Peer benchmark | Highest severity observed |
|---|---|---|---|
| IRB — PD model | 6.2 | 14.0 | F4 |
| IRB — LGD model | 5.8 | 12.5 | F4 |
| IRB — CCF & off-balance-sheet | 3.1 | 7.0 | F3 |
| IRB — data quality & lineage | 4.6 | 11.0 | F4 |
| IRB — model validation & monitoring | 4.9 | 10.5 | F4 |
| Credit risk — NPE classification | 2.3 | 6.2 | F3 |
| Credit risk — IFRS 9 staging & SICR | 3.8 | 9.0 | F4 |
| Counterparty credit risk — MPoR | 1.9 | 4.5 | F3 |
| Governance & documentation | 4.1 | 8.5 | F2 |
| Pillar 2 / ICAAP narrative | 1.7 | 4.0 | F2 |
Supervisory priority themes on live ECB OSI mandates — not what the SSM published in its Priorities document but what the inspectors are actually asking the bank.
Inspectors pushing back harder on downturn LGD adequacy — especially in real-estate-heavy portfolios. The 2008–2014 downturn window is being challenged as insufficient for book risk.
Re-litigation of the 2021 EBA RTS on material changes. Banks are discovering that many "non-material" refreshes were in fact material under the new thresholds.
New theme. Inspectors now asking to see Q1 2026 parallel runs during this year's inspections. Banks that haven't started have nowhere to hide.
The ECB publishes no finding-level statistics. The EBA’s peer-review reports are two years lagged and polite. The large generalist firms have the data but do not publish it — and that is a perfectly rational commercial choice, because unpublished intelligence is priced higher than published intelligence. We come at this from a different angle: as a senior-only CRR3 and OSI specialist, the published benchmark is how we start the conversation.
We publish it. Because once a bank sees that 11 findings on IFRS 9 staging is normal and 4 is possible, the conversation inside the bank changes. That conversation change is what we sell.
The full member-only tracker includes: bank-size-tier breakdown, country-by-country severity distribution, a rolling 3-year trend, and the top 20 "preventable" finding types with our fix playbook.
The OSI Findings Tracker is an anonymised quarterly benchmark built from the subset of ECB/SSM on-site inspections in which Ezelman has held a delivery or readiness mandate role. Individual institutions, individual findings and individual JSTs are not identified — the numbers are aggregated to portfolio theme and severity.
Sample. Rolling 4-quarter window of ECB/SSM on-site inspections supported by Ezelman from Q1 2024 onwards. Only inspections where Ezelman had visibility to the final or draft inspection report are included. “Findings” are inspector-logged observations in those reports, regardless of severity grade.
Peer benchmark. The “Peer benchmark” column is Ezelman’s reconstruction of the market range from (i) the SSM Supervisory Priorities (2024–2026), (ii) EBA peer-review reports and methodological notes, and (iii) public statements by ECB Banking Supervision representatives on inspection-outcome distributions. It is a triangulated anchor, not a published supervisory statistic — no such statistic exists.
Severity. F1–F5 severity labels follow the ECB on-site inspection methodology. F4 is a finding requiring specific supervisory response; F5 is the rare reopening-of-model-approval category.
What we do not do. We do not publish bank names, institution size, JST identity, or specific mandate-level outcomes. Consistent with our sitewide public-data policy: every figure on ezelman.com is either public-sourced, an estimate with stated methodology, or an anonymised mandate-derived benchmark.
Next edition: Q3 2026.