Modelling and Validation

Where we help

Validation is still mostly done by firms who are conflicted (their audit team is signing the financial statements), under-specialised (they understand models but not data pipelines, or vice versa) or under-priced for what is involved (a junior team, a senior signature). Boards and regulators are increasingly intolerant of this.

What we do

• Independent IFRS 9 ECL model audits. Methodology review, code review, reperformance, and the calculation of independent challenger estimates against unsecured and secured credit-risk models. Documentation and communication of procedures, quantitative findings and areas of non-compliance.
• Basel capital model validation. Code review and reperformance of capital calculations for unsecured lending portfolios at Tier 1 banks, with a written validation report.
• Validation of actuarial cashflow models. Reinsurance, premiums, benefits, commission and fees across life, health, general and reinsurance products.
• Independent reproduction where in scope. For cashflow models and credit-risk models we have validated, independent reproduction is part of the standard methodology. For very large estates (e.g. Basel internal-model platforms running into hundreds of thousands of lines), reproduction is scoped to the components material to the validation question.
• Methodology review. Against actuarial standards (ASSA), regulatory requirements (IFRS 17, IFRS 9, SAM, Solvency II) and internal policy.
• Validation of AI-assisted workflows. Explicit testing of explainability, drift and bias on AI-assisted actuarial and financial-services outputs.
• Due diligence on loan books. Analysis and interpretation of key performance metrics using historical data — for acquirers, regulators and audit committees.
• Expert-witness services on actuarial, modelling and data-related disputes.
• Audit-adjacent support that does not create conflicts with statutory audit.

Named frameworks we apply

Two named methodologies we apply on this work:

• The Overlay Taxonomy. A structured classification of the manual overlays sitting between an actuarial calculation engine and a disclosed number — what they are, why they exist, which can be retired, which need to be re-built into the pipeline, which are doing real methodological work that the engine cannot.
• The Four-Extension AI Validation Protocol. Four explicit extensions to standard model validation that AI-assisted financial-services workflows require: explainability testing, drift testing, bias testing, and reproducibility testing under prompt and weight change. The same protocol we apply to AI in our own delivery — see How we use AI.

Both are open methodology — referenced in our Insights articles, free to use with attribution.

Engagement model

Fixed-fee scoped engagements with clear deliverables and timelines. Independent of audit, hyperscaler and software-vendor incentives. Senior actuarial, banking and engineering reviewers — not delegated to junior staff. Independent reproduction is part of the methodology where it is in scope — see above on how reproduction is scoped on very large model estates.