G/00 // Guide — Explainability

How do lenders produce explainable adverse-action notices?

Explainable adverse-action notices are produced from monotonic-constrained models plus feature-attribution (SHAP or equivalent), translated into a plain-English CONC-compliant template. The output is the top three contributing features, an intelligible reason and — where honest — a suggested corrective path. This is a Consumer Duty accelerator, not a compliance cost.

Short answer

Explainable adverse-action notices are produced from monotonic-constrained models plus feature-attribution (SHAP or equivalent), translated into a plain-English CONC-compliant template. The output is the top three contributing features, an intelligible reason and — where honest — a suggested corrective path. This is a Consumer Duty accelerator, not a compliance cost.

Why generic notices fail Consumer Duty

A boilerplate 'we are unable to offer credit at this time' notice is technically compliant and functionally useless. Consumer Duty raises the standard: outcomes matter, and an outcome that leaves a customer no clearer about what to do next is a poor outcome.

The mechanics of an explainable notice

  1. Model architecture with monotonic constraints on the features that matter most.
  2. Per-decision feature attribution (SHAP or equivalent).
  3. Feature-to-language mapping curated by compliance and product together.
  4. CONC-compliant template with plain-English top-three reasons.
  5. Corrective-path suggestion where honest — omitted where not.
  6. Full audit trail: model version, features, attribution scores, template version.

Related questions

Is SHAP acceptable to the FCA?

Yes, when combined with monotonic constraints and a documented governance package. It is the current practitioner default.

What about black-box models?

The architecture assumes explainability constraints. A model that cannot be explained is not shipped to a customer-facing decision.

Corrective-path suggestions — legal risk?

Only suggest what is genuinely likely to help. Legal reviews the language templates once; per-decision rendering is automatic.

Multilingual?

Yes, at template level.

Timeline?

Six to ten weeks after the £15k Diagnostic for the explainability layer over an existing model.

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.