Description

A focused course, tailored for you

The SR 11-7 ML Model Validation Playbook

Bring machine-learning models into the existing model risk lifecycle without rewriting your inventory, your validation memo template, or your committee charter.

The Model Risk Committee deck has an ML model on it and the validation template was written for logistic regression. The memo has to read clean to the OCC and the model has to clear conceptual soundness, outcomes analysis, and ongoing monitoring under SR 11-7. Nobody on the team has done this end to end yet.

$199 one-time

Tailored to your situation. Access within 24 hours. 30-day money-back.

Includes a hand-built implementation playbook delivered alongside course access, generated for your specific situation.

Why this course

Risk analytics managers at large broker-dealers and banks are sitting on a queue of ML models the business wants in production. The model risk function inherited an SR 11-7 lifecycle, a validation memo template, an inventory taxonomy, and a committee cadence that all assume linear or logistic models with stable populations. ML breaks every assumption underneath those artefacts. Conceptual soundness has to handle feature engineering decisions and target leakage. Developmental evidence has to cover hyperparameter tuning provenance, not just coefficient estimates. Outcomes analysis has to handle non-stationary populations and the validator has to decide whether champion-challenger is still the right back-test design. Ongoing monitoring needs PSI on input distributions, drift detection on outputs, and a documented re-fit trigger that the committee will approve. The OCC and Fed examiners are reading these memos with the same SR 11-7 lens, so the memo has to translate ML mechanics into the conceptual soundness, process verification, and outcomes-analysis structure they expect. The course walks the rebuild section by section, with worked memos and the back-test designs that survive validation review.

What you walk away with

Rewrite the four sections of an SR 11-7 validation memo so ML models clear committee review on first submission.
Design back-tests for tree-based and neural-network models that hold up under MRA or MRIA scrutiny.
Author the ongoing monitoring suite (PSI, drift, re-fit triggers) that the committee will approve as part of the validation finding closeout.
Update the model inventory taxonomy to capture ML-specific risk attributes without breaking the existing tier definitions.
Translate ML mechanics into the conceptual soundness, process verification, and outcomes analysis language examiners expect.

The 12 modules

Module 1. What SR 11-7 Actually Requires of an ML Model

A line-by-line read of the conceptual soundness, process verification, ongoing monitoring, and outcomes analysis pillars, restated for ML. What changes when the model is a gradient-boosted tree versus a logistic regression. The interpretive guidance the OCC and Federal Reserve have issued since the original supervisory letter, and which of it is binding versus illustrative. The lens you will carry through the next eleven modules.

Module 2. The Validation Memo Template, Rebuilt

The four sections of the standard SR 11-7 validation memo, with the ML-specific content each section now has to carry. Conceptual soundness expands to feature engineering and target leakage. Developmental evidence expands to hyperparameter tuning provenance and cross-validation design. The downloadable Word template ships with each section pre-headed and pre-prompted so a validator can populate without re-inventing structure on a per-model basis.

Module 3. Conceptual Soundness for Tree-Based and Neural Models

How to write the conceptual soundness section for a GBM, random forest, or neural network. The narrative arc that walks a non-quant examiner from business problem to model choice to feature set without invoking algorithm-specific math the reader will not follow. Worked examples for a consumer credit GBM and a market-risk neural network, including the diagrams that translate model architecture into examiner-readable form.

Module 4. Developmental Evidence and Tuning Provenance

What developmental evidence looks like when the model has fifty hyperparameters and a tuning grid that ran for two weeks. The artefacts you need to preserve from the development workflow (cross-validation splits, hyperparameter search logs, feature importance traces) and the artefacts the validator does not need to see. The audit trail the committee expects when the developer says the model was retrained on more recent data.

Module 5. Outcomes Analysis and Back-Test Design

The back-test designs that hold up under MRA or MRIA scrutiny. Why champion-challenger is necessary but not sufficient for tree-based models, what walk-forward validation actually proves about a market-risk model, and the population stability assumptions you have to test before you trust any back-test result. Three worked back-test designs with the statistical evidence tables a validator would attach to the memo.

Module 6. Ongoing Monitoring: PSI, Drift, and Re-Fit Triggers

The monitoring suite an ML model needs once it is in production. Population stability indices on each input feature, prediction drift metrics that catch concept drift before performance degrades, and the re-fit trigger that the committee approves at validation time so the developer is not negotiating it during a stress event. The dashboard layout and threshold-setting logic that satisfies the ongoing-monitoring pillar.

Module 7. The Model Inventory Entry for ML Models

How to extend the existing model inventory taxonomy to capture ML-specific attributes (training data lineage, feature pipeline owner, retraining cadence, MLOps platform of record) without breaking the tier definitions the committee already uses. The inventory schema additions, the validation-status fields, and the dependency mapping that lets the bank answer an OCC inventory request without a six-week data pull.

Module 8. Handling Champion-Challenger and Benchmark Models

When the regulator asks for a benchmark model and the only benchmark that runs in production is the legacy logistic regression. How to design a credible challenger that is simpler than the champion, fast to retrain, and gives the committee a defensible reference point. Worked challenger specifications for credit, market, and operational risk use cases, and the comparison table that goes into the memo.

Module 9. Explainability Evidence That Holds Up at the Committee

SHAP, LIME, partial dependence, accumulated local effects: which explainability artefacts the Model Risk Committee will accept as evidence of conceptual soundness and which will be sent back. The trap of post-hoc explanations that contradict the conceptual narrative. The explainability appendix the validator attaches and the executive summary version that goes in the front of the memo.

Module 10. Closing MRA and MRIA Findings on ML Models

The structure of a Matter Requiring Attention or Matter Requiring Immediate Attention written against an ML model, and the remediation plan that closes it. What the OCC and Federal Reserve examination teams actually accept as evidence of closure: documented monitoring thresholds, a re-validated memo, or a re-developed model. The closeout package and the timeline the committee can defend.

Module 11. Governance: Committee Charter, Tiering, and Approval Authority

How the committee charter, the tiering policy, and the approval-authority matrix need to update once ML models are in scope. The tier-1 versus tier-2 thresholds that no longer make sense when a credit model touches every retail account, the approval delegation that has to be re-drawn, and the standing-agenda items the committee adds (drift report, re-fit notifications, vendor-model attestations) so ML governance is visible at every meeting.

Module 12. The 90-Day Stand-Up Plan for Your Function

A day-by-day plan to move your function from one ML model on the agenda to ten in the inventory cleanly. Week one is the inventory schema update. Week two is the memo template rollout. Week three is the back-test design library. Week four is the monitoring dashboard build. Each week ships an artefact the committee or the OCC can see, so the function is visibly moving and the next ML approval lands on a built lifecycle, not an ad-hoc one.

How this addresses your situation

Specific modules that map to what you said you are dealing with.

You have an ML model on this quarter's Model Risk Committee agenda and the validation template needs a rebuild before the memo goes up.

You are responding to an MRA finding against an existing ML model and the closeout package has to land before the next exam cycle.

Your inventory is about to be requested by the OCC or Federal Reserve and the ML model entries are thin, inconsistent, or missing.

Your function is hiring or onboarding validators who need a structured playbook to come up to speed on ML model validation.

What you get with this course

Twelve text-based modules covering the SR 11-7 lifecycle restated for ML.
Downloadable validation memo Word template with all four sections pre-structured.
Three worked validation memos: a GBM consumer credit model, a neural-network market-risk model, and an operational-risk anomaly detection model.
Back-test design library with statistical evidence templates for each design.
Ongoing monitoring dashboard layout with PSI, drift, and re-fit trigger specifications.
Model inventory schema extension with ML-specific fields and dependency mapping.
Validator checklist and committee-deck template for the approval submission.
Hand-built implementation playbook tailored to your function, your model mix, and your committee cadence.

What you will have in hand by Day 1, Week 1, Month 1

Account provisioned in the Art of Service learning environment within 24 hours of purchase, with the hand-built implementation playbook delivered alongside it.

Twelve modules available immediately on provisioning, with downloadable templates and worked examples for every module.

Hand-built implementation playbook tailored to your function and your model mix delivered as a single document alongside course access.

Lifetime access to the modules and template library. Re-download as your inventory grows.

Before and after

Before

An ML model on the agenda, a validation memo template that was written for logistic regression, an inventory that does not capture ML-specific risk attributes, and a Model Risk Committee that has to interpret each ML submission on first principles.

After

A rebuilt SR 11-7 lifecycle that handles ML cleanly. Memo template, back-test designs, ongoing monitoring suite, inventory schema, and committee tiering all updated. New ML models land on a built lifecycle and clear committee review on first submission.

What happens if you do not address this

Each ML model approved on an ad-hoc validation process is a future MRA finding. Each ML model deployed without a documented re-fit trigger is a future stress-event surprise. Each ML model entered into the inventory without ML-specific attributes is a gap the OCC or Federal Reserve will find on the next exam. The cost of doing the rebuild now is one course and one playbook. The cost of doing it after an MRA is a remediation programme, an exam memo, and a 12-month committee-visibility tax.

Who it is for

Senior Manager or Director in Model Risk Management, Risk Analytics, or Model Validation at a large broker-dealer, bank, or wealth manager. You own one or more model lifecycle stages under SR 11-7 or FRB SR 15-18. You sit on or report to a Model Risk Committee. You have a queue of ML or AI models the business wants in production this fiscal year and the existing validation playbook does not cover them cleanly. You write or review validation memos that need to read defensibly to the OCC, Federal Reserve, or FINRA.

Who this is NOT for. Data scientists building ML models for the first time. Compliance generalists with no quantitative background. Executives looking for a board-level overview of AI risk. This course assumes you already know what a validation memo is, what a back-test is, and what SR 11-7 requires.

How it arrives

Text-based course in the Art of Service learning environment, plus downloadable templates and worked examples for every module, plus the hand-built implementation playbook delivered alongside course access.

Time investment. Eight to twelve hours of reading across the twelve modules. Three to five hours of template adaptation per model the validator runs through. The implementation playbook is a reference document, not a workbook, and is consulted as the function rebuilds its lifecycle artefacts.

Why $199 is the right number

Big-four consulting engagements cover the same ground but start at six figures and ship a slide deck rather than a working memo template. Internal training sessions cover the conceptual material but rarely ship the artefacts (memo template, back-test designs, inventory schema) that a function actually needs. Free SR 11-7 supervisory guidance from the regulators is the source material but does not translate to ML or ship templates. This course is the artefacts plus the translation plus the hand-built playbook, at the price of an industry conference ticket.

FAQ

Does this assume I already know how ML models work?

Yes. The course is for risk-analytics managers and validators who already understand model mechanics. It teaches how to bring those mechanics into the SR 11-7 lifecycle, not how to build ML models from scratch.

Is this only for banks under FRB supervision?

No. The SR 11-7 framework is the de facto standard for model risk management at broker-dealers, wealth managers, and insurance carriers as well. The course covers the FRB lens and notes where OCC, FINRA, and state-insurance regulator expectations diverge.

Do the templates work for my inventory tooling?

The templates are tool-agnostic Word and Excel artefacts. They populate any inventory platform that accepts structured metadata, and the schema extension is documented for direct import into the common GRC and model-inventory platforms.

What does the hand-built implementation playbook actually cover?

It is a per-buyer document built after purchase. It maps the course modules to your function, your model mix, your committee cadence, and your current validation pipeline. Where the course is general, the playbook is specific to your situation.

How current is the regulatory content?

The course tracks the supervisory letters and examination guidance in force at the time of purchase. The implementation playbook explicitly notes any guidance that has moved since the course was last refreshed.

30-day money-back guarantee. If after a week of working through the materials this is not what you needed, reply to the receipt email and a full refund is processed. No questions, no forms.

Within 24 hours your account in the learning environment is provisioned and the tailored implementation playbook is delivered alongside it.