Engagements
The Same Exposure,
Engaged Three Ways.
One body of work — measuring what a model or an agent exposes an institution to, and documenting the method behind the number. What changes is who retains it: counsel building a record, a lender or seller-servicer meeting control and governance requirements, or an audit function building readiness to test it themselves.
Engagement 01
Counsel-Retained Expert Analysis
Strategic Focus Areas
- Algorithmic fairness and Less Discriminatory Alternative analysis — LDA comparability under ECOA, the Fair Housing Act, and employment selection; the alternatives searched and the basis for each rejection
- Disparate impact measurement — selection rate and adverse impact analysis across protected classes; comparison population construction, control-variable specification, and proxy review of what a variable stands in for
- Re-identification risk — quasi-identifier analysis, attacker modeling, and statistical disclosure risk in data and models, against HIPAA's “very small” standard and GDPR's “reasonably likely” means test
- Whether a model performed as represented — measured against what was claimed to a board, an examiner, a counterparty, or a buyer; model reconstruction and decision attribution where the record is contested
Representative Deliverables
- Expert report stating method, populations, effect sizes, assumptions, and limitations
- Less discriminatory alternative analysis with the alternatives searched and the basis for each rejection
- Signed Expert Determination or anonymisation assessment, with assumptions, controls, and validity period
- Model reconstruction and decision attribution report
- Consulting-expert analysis under privilege; testifying-expert report and disclosure where the matter calls for it
- Deposition and cross-examination support
ECOA / Reg BFair Housing ActTitle VII / UGESPHIPAAGDPR
Representative Engagement
- Validated and audited underwriting, credit, prepayment, fraud, and employment screening models at Fannie Mae under continuous FHFA examination, including the fair lending and disparate impact exposure those models carry.
- Audited approximately 200 models across credit, allowance, fair value and AI/ML applications under continuous federal examination, testing methodology, implementation, and outcome performance.
Engagement 02
Lender and Seller-Servicer Advisory
Strategic Focus Areas
- Model and AI system inventory — completeness, classification, and materiality tiering
- Designated-owner assignment and the annual review cycle Fannie Mae LL-2026-04 and Freddie Mac Guide 1302.8 require
- Model risk policy and validation standards — what each tier requires, who may perform it, and what evidence is retained
- Committee charters, escalation paths, and board reporting
- Second-line role and authority design, including independence from model development
- Vendor, subservicer, and technology-provider governance — diligence, contractual evidence rights, and ongoing oversight
- Technology providers: the AI governance documentation lender clients now require, and the evidence behind each representation
- Disclosure readiness: what must be represented to the agencies, and the evidence behind each representation
- Conceptual soundness, outcome and performance testing, benchmarking, and drift assessment for credit, prepayment, and servicing models
- NIST AI RMF and ISO/IEC 42001 as delivery vocabulary, mapped to the agency requirement rather than substituted for it
Representative Deliverables
- Model and AI system inventory with classification and materiality tiering
- Model risk management policy and validation standard
- Annual review template and calendar keyed to designated owners
- Committee charter, escalation path, and board reporting pack template
- Vendor and subservicer governance standard with evidence requirements
- Validation report with quantitative acceptance criteria, testing results, and stated limitations
- Gap assessment against the agency requirements, with remediation guidance
- Examination response package — inventory, validation status, and control evidence
Model InventoryValidation StandardsAnnual ReviewVendor & Subservicer Governance
Representative Engagement
- Owned the firm-wide model definition, risk tiering methodology, and validation standard at a Tier-1 U.S. financial institution, covering approximately 1,000 model owners across finance, treasury, wealth management, and compliance.
- Determined which machine learning and artificial intelligence systems met the definition of a model and what controls applied to each.
- Built committee charters, escalation paths, and board reporting for model risk oversight.
Engagement 03
Auditing Agentic AI and Models
Strategic Focus Areas
- Scoping an audit of an AI or ML model — what is in scope, what evidence to request, and what to test
- Reading a validation report critically: what a conceptual soundness finding actually establishes, and what it does not
- Outcome and performance testing in an audit context, and what a result does and does not support
- Fairness testing from the audit chair — what an auditor can conclude, and where the conclusion belongs to counsel instead
- Auditing agentic systems — tool use, chained actions, and decisions taken before a human reviews them
- Evidence standards: what a model audit file must hold to survive examination
- Calibrating audit coverage to model risk ratings
Representative Deliverables
- Cohort delivered in-house to a named audit or risk function
- Audit programme and work-paper templates for model engagements
- Evidence and documentation standards for a model audit file
- Practitioner reference material the function retains
Chief Audit ExecutivesIn-House DeliveryModel Audit MethodEvidence Standards
Representative Engagement
- Delivered model risk training and guidance to model owners across finance, treasury, wealth management, and compliance at a Tier-1 U.S. financial institution, covering approximately 1,000 model owners.
- Built and ran the consultation process for contested model classification cases, then assessed whether the standard was being applied correctly in practice.
- Delivered risk-aligned briefings calibrated to model risk ratings, establishing a shared taxonomy and definitions for AI and ML models across cross-functional teams.
The Approach
Every Engagement Follows the Same
Four-Phase Discipline
01
Scope
The matter, the model population, or the cohort defined at intake — what is in scope, what data and documentation are required, what the deliverable is, and the date. For an expert matter, the side is confirmed before the first substantive call. Approved before work begins.
02
Measure
Testing and analysis against the standard that actually applies — the legal test, the agency requirement, or the audit standard. Findings carry the method, the population, and the effect size behind them.
03
Document
A written product stating method, results, limitations, and the basis for each finding — written to be handed to a court, an examiner, a board, or an audit committee without translation.
04
Transfer
Protocols, templates, and thresholds your own team operates afterwards. Clean disengagement.