## FCA SS1/23, PRA Expectations and the SR 11-7 Benchmark
UK model risk management for banking now centres on FCA Supervisory Statement SS1/23 and its PRA sister guidance, both of which draw heavily on the conceptual framework first established by the US Federal Reserve and OCC's SR 11-7 guidance in 2011 — still the most widely referenced international benchmark for model risk management maturity, even for UK institutions with no US regulatory footprint. SR 11-7's core contribution was formalising model risk as a distinct risk category requiring board-level ownership, not merely a technical concern for quantitative teams, and defining the now-standard three lines of defence for models: model development (first line), independent model validation (second line), and internal audit (third line). UK regulators have adapted rather than copied this framework — SS1/23 places explicit emphasis on proportionality (smaller building societies face lighter documentation expectations than G-SIBs) and on ML-specific risks that SR 11-7's 2011 vintage did not originally anticipate, such as training data drift, explainability for credit decisioning under the Equality Act 2010, and LLM hallucination risk in customer-facing or risk-function applications.
## Model Inventory: The Foundation of Defensible Governance
Every UK bank subject to SS1/23 must maintain a comprehensive model inventory — a single source of truth cataloguing every model in production use, whether a simple logistic regression credit scorecard or a deep learning fraud detection system. A defensible inventory entry captures: model purpose and business owner, materiality tier (typically high/medium/low based on financial or customer impact), development and validation dates, current validation status, known limitations, and downstream model dependencies (many banks discover during inventory-building exercises that a 'model' is actually a chain of three or four interdependent models feeding one another, a common source of underestimated aggregate risk). UK banks report that the initial inventory-building exercise is frequently the single largest effort in a model risk management uplift programme, since undocumented 'shadow' models built by individual analysts outside formal model development processes are a near-universal finding — often 20-40% of models discovered in an initial inventory sweep were previously unknown to central risk functions.
## Independent Validation for Credit, AML and Trading Models
Independent Model Validation (IMV) must sit organisationally separate from model development, reporting through a distinct risk chain to avoid the conflict of interest inherent in developers validating their own work. For credit risk models, IMV assesses discriminatory power (Gini coefficient, KS statistic), calibration accuracy (comparing predicted versus observed default rates via backtesting), and fairness across protected characteristics under the Equality Act 2010 — a UK-specific validation dimension with no direct SR 11-7 equivalent, reflecting the FCA's Consumer Duty emphasis on avoiding foreseeable harm to vulnerable customer groups. For AML transaction monitoring models, validation must additionally assess false positive rates against SAR (Suspicious Activity Report) quality — a model tuned purely to minimise missed detections without regard to false positive volume creates its own operational risk by burying genuine alerts in noise, a failure pattern the FCA has specifically flagged in AML model reviews. For trading and market risk models, independent validation includes stress testing against historical crisis scenarios and P&L attribution analysis to detect model risk masquerading as trading performance.
## Ongoing Monitoring: PSI, CSI and Drift Detection
Model validation is not a point-in-time exercise; SS1/23 and PRA guidance both expect continuous ongoing monitoring proportionate to model materiality. Standard monitoring metrics include Population Stability Index (PSI) tracking whether the population a model scores has drifted from its training population, Characteristic Stability Index (CSI) tracking individual feature distribution drift, and performance decay tracking (Gini/AUC trend over time against a defined tolerance band). High-materiality models — retail credit scorecards, AML transaction monitoring, trading risk models — typically require monthly monitoring dashboards reviewed by the model owner with escalation triggers defined in advance (e.g., a PSI above 0.25 triggers mandatory re-validation review within 30 days). UK banks increasingly automate this monitoring through MLOps platforms that generate monitoring evidence directly consumable by validation and audit functions, since manually compiled monitoring packs are both error-prone and a common finding in PRA thematic reviews.
## Building a UK-Compliant MRM Programme
A realistic uplift programme for a mid-size UK bank or building society bringing an existing model estate into SS1/23 compliance typically runs 6-9 months for the full model inventory, initial validation backlog clearance and monitoring dashboard build, followed by an ongoing steady-state operating rhythm. The most common failure mode in UK MRM programmes is treating documentation as a one-off compliance exercise rather than embedding it into the model development lifecycle from the outset — banks that require a completed Model Purpose Document and validation plan before a model reaches production, rather than retrofitting documentation afterward, spend materially less on remediation over time. Codeworth's machine learning engagements supporting UK banks and building societies on model risk management typically combine an initial gap assessment against SS1/23 and PRA expectations with hands-on build support for the model inventory and monitoring infrastructure, since most institutions have the governance policy documents but lack the underlying tooling to operationalise them.
Related Codeworth Services
Need Help with Implementation?
The Codeworth team is ready to implement these ideas in your project. Free consultation.
Discuss the ProjectCodeworth team specialist.