Back to Articles

Model Inventory in Banking: What to Record and How to Keep It Current

11/15/2026
18 min read
Model Inventory in Banking: What to Record and How to Keep It Current

A banking model inventory is the single, maintained record of every model a bank or credit union has in development, in use, or recently retired, with enough detail to understand model risk for each model and across the whole institution. Regulators describe maintaining one as common industry practice, and the 2026 interagency guidance says an effective inventory supports an understanding of model risk at both the individual and aggregate levels, according to the SR 26-2 guidance. This article covers what belongs in the inventory, which fields to capture, how to tier models, and how to keep the record accurate over time.

TL;DR

  • The inventory is the foundation of model risk management: you cannot validate, monitor, or tier what you have not listed.
  • SR 26-2, issued April 17, 2026, supersedes SR 11-7 and keeps the inventory expectation, with oversight scaled to the size and complexity of the institution.
  • Record purpose, owner, inputs, outputs, vendor status, risk tier, validation dates, known limitations, and policy exceptions for each model.
  • Tier every model, store the scores and rationale with the date, and let the tier set validation depth and frequency.
  • Treat the inventory as a living control: update it on every model change, new use, or monitoring breach, not only at year end.
Model Governance

Automate Regulatory Model Risk Governance

Examine models against 32 qualitative criteria and resolve risk Tiers with pre-deployment checklists per OCC 2011-12 guidelines.

Table of Contents

What a model inventory is and why examiners ask for it

The inventory is a register, not a policy. The policy says how the institution governs models; the inventory shows what is actually being governed. When an examiner asks which models drive your allowance, your underwriting decisions, your BSA/AML alerts, or your stress results, the inventory is the document that answers in one place.

The guidance frames it as a comprehensive set of information for models under development or in use that supports effective model risk management. The FDIC's earlier statement of the same expectation is specific about lifecycle coverage: models implemented for use, under development for implementation, or recently retired should all be tracked. Retired models matter because their outputs may still sit inside historical reports, and because an examiner may ask why a model was withdrawn.

For risk managers, the inventory does three jobs at once:

  1. Scoping. It defines the population that validation, monitoring, and reporting have to cover.
  2. Prioritization. With risk tiers attached, it tells you where limited validation capacity goes first.
  3. Aggregation. It lets management and the board see model risk in total, including dependencies where one model feeds another.

The third job is the one most institutions under-build. A list of model names does not show that your CECL estimate depends on a prepayment model, which depends on a data feed from your core system. Dependencies belong in the record.

Which guidance applies to banks and credit unions

The current federal standard is SR 26-2, the revised model risk management guidance issued jointly by the Federal Reserve, the OCC, and the FDIC. The Federal Reserve's letter states that it supersedes and replaces SR 11-7 and the 2021 interagency statement on model risk management for BSA/AML systems. The OCC published the same guidance as Bulletin 2026-13A, and the FDIC issued earlier versions through FIL-17-2022.

Three points matter for inventory design.

First, proportionality is explicit. The OCC bulletin states that the nature and sophistication of model governance is typically informed by the extent and sophistication of model usage as well as the size and complexity of the organization. A community bank with a few dozen models should not carry the inventory apparatus of a large dealer bank, but it still needs a complete and current list.

Second, the guidance is supervisory guidance, not a rule. That does not make the inventory optional in practice. Examiners test model governance against it, and a missing or stale inventory is usually the first visible weakness. For a deeper look at what changed from the 2011 framework, see our overview of SR 26-2 and why it reshapes model risk management.

Third, credit unions should confirm with their examiner how model governance expectations apply to them. The interagency guidance is issued by the three banking agencies, but the logic of an inventory, tiering, validation, and documentation is the same sound practice any lender relying on quantitative estimates should be able to demonstrate. Building to this standard prepares a credit union for whatever its own regulator asks.

What belongs in the inventory: defining a model

Scope is the first decision, and it is where inventories most often go wrong, either by listing every spreadsheet or by missing the vendor tools that carry the most risk.

SR 26-2 defines a model as a complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates. It excludes simple arithmetic calculations and deterministic rule-based processes, and software without those theories underpinning its design or use.

Applied to a bank or credit union, that usually puts the following on the list:

  • Credit scoring and underwriting models, including application scorecards and behavioral scorecards
  • Allowance and CECL models, including loss-rate, discounted cash flow, and PD/LGD approaches
  • Stress testing and capital planning models
  • Interest rate risk and asset-liability management models, including deposit decay and prepayment assumptions
  • Liquidity forecasting models
  • BSA/AML transaction monitoring and customer risk-rating models
  • Fraud detection models
  • Pricing and profitability models

The harder cases are tools near the boundary. A rule-based decision matrix may fall outside the formal definition, yet the OCC's handbook notes that some quantitative tools may not meet the definition of a model in the guidance, while sound risk management typically includes maintaining an inventory of all quantitative tools. A practical approach is a two-part register: a formal model inventory, and a companion list of quantitative tools and end-user computing files that were assessed and judged not to be models, with the reason recorded. That second list answers the examiner's inevitable question about what you decided to leave out.

Vendor models

Vendor and third-party models belong in the inventory with the same discipline as internal ones. The guidance observes that the widespread use of customized vendor products, including data, parameter values, or complete models, can present unique challenges for validation, and it identifies validation of vendor products, by internal or outside parties, as an important element of model risk management. Many community institutions run most of their quantitative estimates through vendor systems, which means the inventory is often mostly a vendor inventory. Record what the vendor supplied, what your institution configured, and what documentation you hold.

Artificial intelligence tools

AI deserves a deliberate decision rather than a default. The guidance states that generative AI and agentic AI models are explicitly excluded from the scope of SR 26-2, while also stating that an institution's risk management and governance practices should guide the appropriate governance and controls for tools and systems it does not cover. The sensible reading is that such tools do not disappear from your governance. Many institutions keep them in the inventory, or in a clearly linked AI register, with an owner, a documented use, input data sources, and a review schedule. Traditional machine learning credit models that fit the definition of a model stay in the formal inventory.

The fields to record for each model

The OCC's handbook lists what a comprehensive inventory may include. The list is long, and not every institution needs every field on day one. The table below groups those fields by the question each answers and flags the ones we would treat as the minimum for a community bank or credit union.

QuestionFieldsMinimum?
What is it?Model identifier, version, description of technology or approachYes
What is it for?Purpose and products supported, actual or expected usage, restrictions on useYes
Who is responsible?Model owner and users (by title or group), developer, validatorYes
Where did it come from?In-house or third party, vendor name, dependencies on other modelsYes
What goes in and out?Type and source of inputs, outputs and their intended useYes
How risky is it?Risk type and level, tier, scoring rationaleYes
Is it sound?Completed and planned validation dates, monitoring frequency, validation results and status, whether it is functioning properlyYes
What is wrong with it?Known issues and limitations, issue status, overlaysYes
What changed?Approval date, last update date, status (development, in use, retired)Yes
What is the exception?Policy exceptions, the time frame the model is expected to remain validRecommended

The FDIC's version of the expectation emphasizes the same core: purpose and products, actual or expected usage, restrictions on use, type and source of inputs, outputs and intended use, whether the model is functioning properly, when it was last updated, and any exceptions to policy.

Two fields deserve extra attention because they are the ones examiners probe.

Restrictions on use. A model validated for consumer auto loans should not quietly start scoring small-business credits. Recording the approved scope turns that drift into something you can detect.

Overlays and limitations. Management overlays, adjustments applied on top of model output, are frequently undocumented. If an overlay materially changes your allowance, it belongs in the inventory next to the model it adjusts.

Tiering: turning a list into a priority map

An inventory with no risk ranking treats a pricing spreadsheet and your allowance model as equals. Tiering fixes that. The OCC bulletin describes effective policies as ones that establish a framework for assessing the magnitude of model risk and applying practices commensurate with that risk.

Research on industry practice shows how firms actually do it. A study in the Journal of Risk Model Validation found that most financial firms use explicit procedures and decision tools to assign models to risk tiers, and that nearly all assign models to several tiers rather than ranking them one by one. Scorecards are more common than decision trees. The same authors note that these tools mainly systematize judgment rather than quantify risk, and that calibration is largely judgmental. In plain terms: the tool gives consistency, and documented judgment gives it meaning.

A workable community-bank scheme uses two dimensions:

  • Materiality: how large the exposure or business impact is, and what the model is used for. Models used for regulatory requirements or to manage financial risk rank higher.
  • Complexity: how sophisticated the method, how opaque the inputs, and how much reliance sits on unvalidated vendor components.

One published course framework maps these scores in a 3x3 matrix to tiers, with Tier 1 receiving the deepest validation, most frequent review, and closest governance. The same source describes a floor rule, under which any model whose output enters regulatory capital or regulatory reporting is at least Tier 2 whatever the matrix says, and recommends storing every tier with its scores, rationale, and date. That last habit is cheap and pays off at exam time, because a reviewer can see why a model sits where it does.

If you want a quick starting point, our OCC 2011-12 model risk tier calculator scores a model on a few inputs, and our write-up on moving beyond guesswork in model tiering explains the logic.

Keeping the inventory current across the model lifecycle

Most inventories fail on maintenance, not construction. A snapshot taken for last year's exam is a liability by the next one. Governance has to name who updates what and when.

The OCC bulletin says sound governance delineates the individuals responsible for key activities throughout the model lifecycle, from development through validation and ongoing monitoring. Translate that into the inventory with a named inventory owner, usually in the second line, and a named model owner for every entry.

The guidance also ties documentation to practical outcomes: it supports continuity of operations, tracking of recommendations, responses, and exceptions, and model remediation. The inventory is where those threads meet, so open validation findings and their status should be visible in it, or linked from it.

Update triggers

Set events that force an inventory update rather than waiting for an annual attestation:

  • A model is approved, changed, or retired
  • A model is used for a new product, segment, or decision
  • A vendor releases a new version or changes its methodology
  • Monitoring shows a breach of a performance threshold
  • A validation finding is opened or closed
  • A tier changes

The Risk Hub course material gives illustrative validation cycles of every year for Tier 1, every two years for Tier 2, and every three years for Tier 3, and states that events such as a model change, a scope extension, a monitoring breach, or a new use trigger validation whenever they happen. Treat those cycles as an example, not a rule; the right frequency is the one your policy sets and you can defend by risk. A tier change should re-date the next validation.

A periodic completeness check

Trigger-based updates catch known changes. They miss models nobody told the inventory owner about. Once a year, run a completeness sweep: ask each business line to list the quantitative tools behind its decisions, reconcile against procurement and vendor lists, and compare with the allowance, capital, and BSA/AML process maps. Each gap becomes either a new inventory entry or a documented decision that the tool is not a model.

Steps in an annual inventory completeness sweep

Common weaknesses examiners find

These patterns recur, and each has a simple countermeasure.

Vendor models listed by product name only. The entry says the vendor system's name but not which estimates it produces. Record the specific outputs your institution relies on.

Ownership by committee. If the owner field says a committee, no one is accountable. Name a title or group that can actually answer questions.

Tier without a rationale. A "High" label with no scores cannot be defended. Store the inputs and the date.

Validation dates that only look forward. Record completed validations and their results, not just the next scheduled date.

Common weaknesses examiners find

Spreadsheet models absent. End-user computing files that feed the allowance or the board's capital reporting often meet the definition of a model or at least belong on the quantitative tools list.

For a wider view of how the inventory sits inside the overall framework, our practical guide to model risk management walks through development, validation, and governance together.

Our perspective

We see the model inventory as the point where model risk management either becomes operational or stays a policy document. Institutions rarely fail because they lack a policy. They fail because the inventory is a spreadsheet maintained by one person, updated before exams, and disconnected from the workflows where models are actually used.

Our view is that the inventory should live next to the work. When underwriting, credit memo drafting, regulatory monitoring, and portfolio review run through governed, auditable workflows with granular audit trails, the evidence an examiner asks for is produced as a by-product of daily activity rather than assembled in a scramble. That is the design principle behind our platform, which is built for credit unions and community banks that need security and compliance without a large model risk team.

We also think tiering should be consistent and written down. A tool that scores materiality and complexity the same way every time, and records the rationale, removes the most common argument in validation planning: why this model and not that one.

Finally, we would urge risk leaders not to wait for perfect scope. Start with the models behind your allowance, underwriting, and BSA/AML, tier them, record owners and limits, and widen from there. A current, honest inventory of twenty models beats an elaborate one of two hundred that nobody trusts.

Raj

Where RiskInMind fits

If your team is spending more time assembling model, compliance, and credit documentation than analyzing it, the work around the inventory is where automation helps most. Our platform is designed to bring regulatory monitoring, credit memo generation, document creation, and audit-ready reporting into one governed environment, with SOC 2 certification and bank-grade security, so the records your inventory points to are consistent and traceable.

Riskinmind

To see how that supports regulatory change tracking, look at Erina, our AI regulatory agent, or review the plans and tiers to find the fit for your institution.

Building the inventory step by step

For an institution starting from a partial list, this sequence works.

  1. Set the definition. Adopt the SR 26-2 definition of a model in your policy, and add a rule for borderline quantitative tools and AI systems.
  2. Harvest candidates. Interview each business line, review vendor contracts, and scan process maps for the allowance, ALM, liquidity, BSA/AML, fraud, and underwriting.
  3. Assess each candidate. Decide model or not a model, and record the reason either way.
  4. Capture the minimum fields. Use the table above, and assign an owner to every entry.
  5. Tier. Score materiality and complexity, apply floors for regulatory use, and save scores, rationale, and date.
  6. Link validation and issues. Connect each model to its last validation, open findings, and next due date.
  7. Set triggers and a review cadence. Define events that force updates, and schedule an annual completeness sweep.
  8. Report. Give senior management and the board an aggregate view: models by tier, overdue validations, open high-severity findings, and vendor concentration.

Credit risk teams can connect the inventory to the downstream controls it supports, such as dashboards and portfolio monitoring; our guide to credit risk dashboard widgets and governance steps covers that link, and our risk assessment in lending guide shows where underwriting models fit in the lending process.

FAQ

How many models should a community bank or credit union expect to list?

There is no standard number, because it depends on products, vendors, and how many estimates run through spreadsheets. Many smaller institutions find a few dozen entries once they include vendor systems for the allowance, ALM, BSA/AML, and underwriting. Completeness and accuracy matter more than the count.

Does SR 26-2 require an inventory?

SR 26-2 is supervisory guidance rather than a rule, but it describes maintaining a comprehensive set of information on models as common industry practice and ties it to understanding model risk at the individual and aggregate levels. Examiners test against it, so treat the inventory as an expected control.

Do vendor models go in the inventory?

Yes. The guidance highlights that vendor products can create validation challenges and says validating them, by internal or outside parties, is an important element of model risk management. Record what the vendor provides, how you configured it, and what documentation you hold.

Are generative AI tools part of the model inventory?

SR 26-2 excludes generative and agentic AI from its scope but says your own risk management and governance practices should guide controls for tools it does not cover. Many institutions track these tools in the inventory or a linked AI register with an owner, a documented use, and a review schedule.

How often should the inventory be updated?

Update it whenever a model is approved, changed, retired, used for something new, or fails monitoring, and run a full completeness check at least annually. Validation frequency should follow each model's tier as set in your policy.

Sources

Recommended

model inventory banking