G/00 // Guide — Treasury

How do lenders use AI in IFRS 9 ECL modelling?

AI improves IFRS 9 ECL modelling by tightening PD estimation per cohort, LGD by segment and EAD projections under forbearance scenarios — with model changes governed under a PRA-comfortable model-risk framework. The wins are per-cohort granularity and speed-of-restatement, not headline provision reduction.

Short answer

AI improves IFRS 9 ECL modelling by tightening PD estimation per cohort, LGD by segment and EAD projections under forbearance scenarios — with model changes governed under a PRA-comfortable model-risk framework. The wins are per-cohort granularity and speed-of-restatement, not headline provision reduction.

What AI actually changes

IFRS 9 ECL has always been model-driven. AI does not replace the model; it improves the PD, LGD and EAD components at cohort level and lets treasury re-run scenarios in hours rather than a quarter. The provision headline may or may not move, but the confidence interval tightens and the audit trail improves.

Where AI adds real value

  • PD segmentation at finer granularity (product × cohort × macro).
  • LGD estimation using loan-level behavioural features (portal engagement, cure history).
  • EAD under forbearance / restructure scenarios.
  • Macro overlay simulation — 100+ scenarios in an afternoon, not a quarter.
  • Backtesting harnesses that satisfy the PRA on model performance monitoring.

Governance is the deliverable

The PRA cares about model risk management. Every AI-touched component ships with the same evidence pack a traditional model would ship with — plus a challenger loop, plus explainability. Treasury and risk sign off on a governance package, not a model.

Related questions

Does the PRA accept AI in ECL?

Yes, with the same model-risk governance any other statistical model requires — plus explainability.

Do we replace our vendor ECL engine?

Rarely. The AI-derived components feed the existing engine; the engine remains the calculation of record.

How does the auditor see it?

Auditors see the governance package, backtests and challenger evidence. This is a normal audit conversation.

Which cohorts benefit most?

Products with rich behavioural data — consumer instalment, SME term, BNPL — see the biggest per-cohort tightening.

Timeline?

Twelve to twenty weeks depending on scope after the £15k Diagnostic.

The AI-first lenders are pulling away on unit economics, not on rate. Every quarter you defer the architecture is a quarter of compounding disadvantage.

The £15k AI Diagnostic maps your lender stack, prioritises the systems that pay back fastest and produces a costed sequenced build plan.