Consulting ยท Data & AI

Independent model validation & regulatory-ready model risk

Independent model validation, AI audit, and enterprise AI governance aligned to SR 11-7, the EU AI Act, and NIST AI RMF - reducing regulatory risk while accelerating trusted AI adoption.

Request a Consultation See our framework
1Uninventoried2Validated3Governed4Monitored5AssuredGravitas Model Risk FrameworkFrom current state to a governed, real-time capability
The problem

What's at stake

Regulated enterprises must demonstrate that their models are sound, validated, and controlled. As AI adoption accelerates, model risk extends beyond traditional quantitative models to machine-learning and generative systems - expanding a perimeter many teams cannot yet cover.

Business impact

Why it matters

Rigorous model risk management satisfies examiners and internal audit, prevents costly model failures, and lets the business adopt AI without accumulating unmanaged risk.

The deeper problem is that models have proliferated faster than the discipline to govern them. In many institutions models are spread across teams, not fully inventoried, and validated, if at all, at a single point in time. As machine-learning and generative models enter decision-making, the population grows more complex and harder to explain, while the framework meant to control it lags behind.

That gap is where loss and regulatory findings originate. A model that was validated once can drift out of correctness unnoticed, an unvalidated model can carry risk no one has priced, and AI models can sit entirely outside the perimeter of governance. The exposure is invisible until it materialises, which is exactly why a continuous, evidenced capability matters.

Our point of view

How we see this

Our point of view on model risk is that validation and management are not the same thing. A model validated at a point in time can still fail in production as conditions change, so the discipline that actually controls risk is lifecycle governance and ongoing monitoring, not a periodic tick in a box. Point-in-time validation on its own gives a false sense of comfort.

We also argue that AI and machine-learning models belong squarely inside the framework. They carry model risk too, and they now sit under both prudential and AI regulation, so leaving them outside governance is a fast-growing and increasingly visible exposure. The same tiering, validation and monitoring disciplines apply, adapted to how these models behave.

Finally, we treat independence as structural, not aspirational. Validation by the people who built a model is not validation, and independence that depends on goodwill erodes under delivery pressure. Setting the function up so independence is built into how it operates is what keeps the assurance credible when it matters most.

Our framework

Gravitas Model Risk Framework

Model risk is often managed as a validation checklist rather than a lifecycle capability. We use a five-stage framework to show sponsors whether the organisation can identify, validate, monitor and govern its models across their whole life, aligned to SR 11-7, OCC 2011-12 and the emerging AI standards, so the board can rely on the numbers models produce.

1UninventoriedModels unknown, unmanagedMost firms start here2ValidatedPoint-in-time checks3GovernedLifecycle controls4MonitoredOngoing drift detection5AssuredBoard-level confidence

Many institutions sit at Validated: models are checked at a point in time but not governed across their life or monitored for degradation. The value of an engagement is moving model risk from episodic validation to a continuous, assured capability.

The five stages: Uninventoried, Validated, Governed, Monitored, Assured. Where your estate sits, what each stage costs you, and what moving up one actually involves, are set out stage by stage in the brochure.

How we work

What a typical engagement looks like

Banking

Situation A growing model population, including machine-learning models, was validated only at points in time.

What we do We build a risk-tiered inventory, strengthen independent validation, and add lifecycle monitoring.

What changes Model risk becomes a continuous capability that satisfies regulators and gives the board evidenced assurance.

Insurance

Situation Machine-learning models sat outside the traditional model risk framework as regulation tightened.

What we do We extend the framework to AI and machine-learning models, mapping controls to SR 11-7 and the EU AI Act, and add drift monitoring.

What changes Degradation that periodic validation misses is caught early, and AI models come inside the governed perimeter.

Asset management

Situation Models were spread across teams with uneven control and high key-person dependency.

What we do We establish a single inventory and independent validation.

What changes Key-person dependency falls and leadership gets a clear view of where model risk actually sits.

FAQ

Frequently asked questions

Which frameworks do you align to?

SR 11-7 and OCC 2011-12 for model risk, and the EU AI Act, NIST AI RMF, and ISO 42001/23894 for AI governance.

Do you validate machine-learning and GenAI models?

Yes. Our framework covers traditional quantitative models as well as machine-learning and generative systems.

Can you provide ongoing validation?

Yes. MRM-as-a-Service embeds ongoing validation and monitoring into operations rather than one-off reviews.

Brochure

Get the full Model Risk Management brochure

Everything on this page and the 32 sections behind it, in one PDF: the full five-stage diagnostic, the delivery path, what the engagement produces, the pitfalls we design around and the questions sponsors ask. Sent to your inbox.

Model Risk Management for your organization

Scope an engagement with a senior practitioner.