SystemCredit Risk & Portfolio Management

Model Risk Management

Model risk management that watches every model monthly, re-scopes the inventory to SR 26-2 and drafts the validation

See one case, screen by screen ↓
Watch with transcript →

Built forModel risk managersModel validatorsModel ownersChief Risk OfficersInternal auditors

The Problem Today

Validators check each model on a yearly calendar, so a model can fail in one segment for a year while the total looks fine.

Read and written up by hand

  • Documents
  • Monitoring
  • Back-tests
  • Findings
  • New guidance, old inventoryEach entry needs a new scope and a new tier
  • AI tools left to each bankGenerative and agentic tools sit outside the new guidance

The Solution

  1. Each model scored monthlyOn production data, against its own control limits
  2. A breach opens revalidationThe task opens with its trigger, scope and evidence attached
  3. Report drafted from evidenceDocuments read and tests run; each section and finding cited
  4. Two people signValidator and head of model risk sign; findings get owners and dates

Built-in controls

  • Agents propose, a person decides
  • Findings close only on a validator’s review
  • Nothing leaves the inventory without a reason

What You Get

A validation plan timed by materiality, so validator hours go to the models that carry the most risk.

23 → 8
days per Tier 1 validation, from evidence to signed report our estimate
  • Segment drift found between validationsA back-test by segment catches what the portfolio total hides
  • AI tools on their own registerEach with its owner, its controls and who checks its output
  • Inventory re-scoped to SR 26-2Scope and tier proposed per entry, for the committee to ratify
In short

What is Model Risk Management?

Model risk management software keeps a bank’s model inventory, tiers each model by materiality, plans and records independent validation, monitors models in production and tracks findings to closure. Model Risk Management adds agents that score each model monthly on production data, re-scope the inventory, read the evidence and draft the validation report, while people decide and sign.

How to choose AI solutions for lending →

Updated

The problem

A model can fail in one segment for a year before its next validation

A bank runs hundreds of models: probability of default and loss given default, the CECL allowance, stress loss projection, deposit beta, prepayment, fraud and customer risk rating. Each one is validated on a calendar. Between validations, someone reads the quarterly monitoring reports, but a model can lose its edge in one segment for a year while the portfolio total still looks calibrated.

Now the rulebook has changed under a live inventory. SR 26-2 replaced SR 11-7 in April 2026. It narrows what counts as a model, ties validation timing to materiality instead of a default annual cycle, and leaves generative and agentic AI for each bank to govern itself. Each entry built under the 2011 guidance needs a scope decision and a tier that internal audit can test, while the validation plan on the old cycles already asks for more hours than the team has.

Where a full validation’s days goestimated
By hand23 days
With the solution8 days
  • Gathering and reading the evidence5 → 1 d
  • Back-testing, stability and the challenger3 → 1.5 d
  • Writing the report and findings10 → 3 d
  • Review and sign-off2 → 1.5 d
  • Chasing open findings3 → 1 d

Estimated split for a Tier 1 credit model’s annual validation, by hand and with the solution.

How it works

Monitoring, re-scoping and validation drafting, agent by agent

What comes in
Models and evidence inEach model, each month · production scores and outcomes, inventory, development docs, prior findings
Agents at work
Outcomes analystcontrol charts, monthly
Then
Inventory scoperSR 26-2 definition
AI tool registrar
Then
Evidence readerdocuments mapped
Materiality tierer
Then
Validation draftersections and findings
Findings tracker
A person decides
Head of Model Riskdecides and signs; validator co-signs
What comes out
Control chart per model
Revalidation task on a breach
Signed validation report
Re-scoped inventory and AI register
One case, step by step

CRE default model drift, the SR 26-2 re-scope and a signed validation

A sample bank with $41.3 billion in assets, 412 inventory entries and the Model Risk Committee in eight days. Here is one Monday, screen by screen, in the working solution.

A worked example on sample data. Names, IDs and figures on the screens and in this section come from the demo.

  1. 0108:52

    Each model on its own control chart

    Head of model risk management

    After the 06:00 run, the monitor shows 184 models on monthly production data: 0 out of control, 2 trending, 182 in control. The CRE probability of default model is trending: rolling 12-month AUC 0.796, center 0.820, lower control limit 0.795. Then the month’s CRE defaults file arrives.

    “3,084 CRE loans to score. The line has been below center for a year and falling each month since December.”

  2. 0209:15

    Out of control, and a revalidation task opens

    Outcomes analyst

    The new point lands at 0.788, below the lower control limit, and current loan-to-value PSI reaches 0.27 against a 0.25 limit. A revalidation task opens on its own with the trigger, the scope (office recalibration; construction-loan use), the lead validator, a link to the annual validation due later this month, and four pieces of evidence.

    “AUC 0.788 against a 0.795 limit, after a year below center and nine months falling.”

  3. 03One click

    The drift point opens its evidence

    Head of model risk management

    The month’s point shows where the drop comes from. By segment, office has fallen to an AUC of 0.68 from 0.80 at development, while multifamily, retail, industrial and hospitality hold. Office loans defaulted at twice the predicted rate: 16 defaults against 8.1 expected. The office back-test, the latest quarterly monitoring report, the challenger results and the change log sit one click away.

  4. 044 min 10 s

    The whole inventory, re-scoped to SR 26-2

    Inventory scoper · Materiality tierer

    The CRE model is not the only thing due before the committee. In the SR 26-2 re-scope, four agents test all 412 entries against the new definition of a model: 298 stay models, 86 move to the tools and calculations register, 28 to the AI tool register, and 14 borderline spreadsheets go to the head of model risk to decide. Re-tiered by materiality, the 12-month plan falls from 214 validations to 131, and from 34,600 hours to 22,320, inside the team’s 22,400.

    “298 + 86 + 28 = 412 · nothing is deleted: each entry that leaves keeps its owner, history and reason.”

  5. 05Each entry

    Model or not, with the reason written down

    Inventory scoper

    Each entry carries its scope test against SR 26-2 §II and its materiality against §III. For the CRE default model: a complex quantitative method, statistical theory, quantitative estimates, not generative AI, so a model. Exposure of $11.8 billion and a financial-reporting purpose keep it in Tier 1.

    “Logistic regression fitted to loan history: statistical theory turning inputs into default estimates, so a model.”

  6. 062 min 59 s

    The annual validation, drafted from 61 documents

    Evidence reader · Validation drafter

    The lead validator starts from the evidence instead of a blank template. The agents read the 146-page development document, four monitoring reports, the change log, last year’s validation and the user guide, ran 23 of 23 tests, and drafted six report sections with 31 citations and four findings (one high, two medium, one low). The proposed rating is fit for use with conditions; the validator can change any severity.

  7. 07Next

    The back-test a portfolio total would have missed

    Outcomes analyst

    Across all CRE, 48 defaults on 3,084 loans (1.56% observed against 1.34% predicted) sits inside the range. Office does not: 3.65% observed against 1.84% predicted, outside the 99% range, p = 0.008. The draft asks for a governed overlay of +181 bp until the office segment is recalibrated, about $13.7 million added to the CRE allowance.

    “Overall the model looks calibrated; the office segment is not. A test on the total alone would have missed it.”

  8. 08Signed

    Two signatures, and conditions the committee approves

    Head of model risk management

    The head of model risk approves this year’s validation of the CRE default model as fit for use with conditions: the governed office overlay, no construction loans rated with the model until that use is validated, and the office segment recalibrated and re-validated within 90 days of the committee. The signature records what it means, and the lead validator co-signs.

  9. 09Chased

    Each finding to an owner and a date

    Findings tracker

    The four new findings join the tracker with due dates set by severity from the committee date: High in 90 days, Medium in 180, Low in 365. Open findings rise from 63 to 67. Reminders go out on the 7 overdue findings, the oldest 104 days late, and the tracker drafts the two extension requests.

    The findings tracker never closes a finding; a validator does.

  10. 10Committee day

    The committee decides what only it can

    Chief risk officer

    The pack closes the Friday before, with five agenda items and four decisions: ratify the 412 scope decisions and 298 tiers and adopt the 131-validation plan; approve the CRE default model with conditions; rule on 7 overdue findings and 2 extension requests; adopt the AI tool standard v1 with 4 control gaps to close in 30 days.

Who it’s for

Model risk work across the first, second and third lines

The model owner, the validator, the head of model risk, the CRO and internal audit, each on the same Monday.

Head of model risk management
Decides and signs
Before
Runs a validation plan that needs 34,600 hours with a team that has 22,400, and hears about drift at the next annual validation.
Now
Sees each model’s control chart monthly, decides the 14 borderline entries, and signs validations drafted from the evidence.
Lead model validator
Validates the credit models
Before
Reads each development document and monitoring report by hand, then writes each section and finding from a blank template.
Now
Starts from 61 documents read, 23 tests run and four cited findings, and spends the time on judgment and severity.
CRE model owner
Remediates the findings
Before
Gets a long report at year end, and an overdue finding on a +35 bp office overlay that was never documented or governed.
Now
Gets each finding with what fails, the evidence behind it, what the validator needs to close it and the date it is due.
Chief risk officer
Chairs the Model Risk Committee
Before
Asks which models could move the allowance, and whether they are validated, then waits for the answer.
Now
Opens a committee pack with each decision asked of the committee: scope and tiers, conditional use, extensions and the AI tool standard.
Model risk audit lead
Tests the inventory decisions
Before
Has to rebuild why an entry left the inventory or changed tier from emails and meeting notes.
Now
Opens any entry and reads the scope test, the materiality and the reason, with read-only access and each step on the record.
The agents

Seven agents that score, scope, draft and track a model inventory

Models are watched monthly on production data, revalidation opens on drift and the validation is drafted; the Head of Model Risk Management and a validator decide and sign.

Inventory scoper

Tests each inventory entry against the SR 26-2 definition of a model (a complex quantitative method applying statistical, economic or financial theory to produce estimates) and routes simple calculations and AI tools to their own registers.

  • Never deletes an entry
  • Borderline entries always go to a person
  • Cites the passage and the evidence for each decision
Materiality tierer

Proposes a tier from exposure, purpose and inherent risk, and re-plans validation timing by tier.

  • Never sets a tier without stating exposure and purpose
  • Tier 1 changes need the committee
Evidence reader

Reads development and vendor documents, monitoring reports, change logs and prior findings, and maps each to a part of the validation.

  • Never treats a developer’s claim as tested evidence
  • Cites the page for each fact
Outcomes analyst

Scores each model on production data monthly, back-tests calibration, measures discrimination and stability, benchmarks a challenger and estimates the allowance effect on the bank’s own data.

  • Never changes a threshold
  • Each result reproducible from its logged run
Validation drafter

Drafts each report section and the findings with citations, and proposes severity and the overall rating.

  • Never signs
  • Never rates a finding without its evidence
Findings tracker

Tracks each finding’s owner, due date and remediation evidence; opens revalidation tasks on a breach; drafts reminders and extension requests.

  • Never closes a finding; a validator does
AI tool registrar

Records each generative or agentic tool’s use, data, owner, the person who checks it and its controls.

  • Never approves a tool for use
Head of Model Risk Management

Approves scope, tiers and validations; validator co-signs; committee ratifies.

Ask in plain words

Model risk questions answered from the inventory, the tests and the findings

The model risk team asks about any model, test, finding or AI tool in plain words, or tells it what to chase. Answers cite their source and use the sample data.

Why does the CRE default model get a high finding?

The office segment fails its back-test. Over the last four quarters, 16 of 438 office loans defaulted against 8.1 expected, 3.65% against 1.84% predicted, outside the 99% range (p = 0.008), and office AUC fell to 0.68 from 0.80. Across all CRE the model looks fine, so a total-only test missed it. The draft asks for a governed +181 bp overlay until recalibration, about +$13.7M to the allowance.

Remind the owners of the 7 overdue findings

Done. Seven reminders sent, each with what the validator needs to close the finding, from small-business override monitoring (104 days late) to the customer risk threshold (11 days). Two owners asked for extensions last month, so both requests are drafted for the committee.

Why did 86 entries leave the model inventory?

SR 26-2 narrows what a model is and excludes simple arithmetic in spreadsheets and deterministic rule-based processes. The 86 are calculators, fee and budget workbooks and fixed-threshold rules such as the wire monitoring scenarios. They move to the tools register with their owner, history and reason, and are still attested once a year. Separately, 28 generative and agentic tools went to the AI tool register.

Which AI tools have a control gap?

Four of the 28: the data quality remediation agent writes to production without approval; the servicing call summary tool has no sample review against recordings; the loan modification intake agent’s actions are not logged to the servicing record; and the collections letter drafter has no person named to review letters before they are sent. The AI tool standard v1 sets the controls and goes to this month’s committee.

The screens

The model monitor, the SR 26-2 re-scope and the validation workspace

From the working solution, on its sample data. Pick one to see it large.

The model monitorThe selected model’s control chart with center, limits and model changes, this morning’s run, and each watched model below.
A breach opens revalidationThe CRE model falls below its lower control limit; a revalidation task opens with its trigger, scope, validator and evidence.
The drift point’s evidenceThe month’s point against center and limits, AUC by segment against development, and the evidence behind it.
The model inventory412 entries with owner, exposure, SR 26-2 scope, tier, last and next validation, and open findings.
SR 26-2 re-scope and re-tier412 entries tested: 298 stay models, 86 tools and calculations, 28 AI tools, with the tier migration and the new plan.
Each entry’s reasonsThe scope test against the definition of a model, the materiality and the validation dates, for internal audit to read.
The validation workspace61 documents read, 23 tests run, four findings with severity, owner, due date and the evidence behind them.
Outcomes analysisPredicted against observed default by segment with the 99% range: office outside it while the total holds.
Sign the validationThe approver signs with the signature’s meaning, the lead validator co-signs, and the committee approves conditional use.
The findings trackerEach open validation and audit finding by severity, owner and due date, overdue first, on a timeline of due dates.
The AI tool register28 generative and agentic tools: what each does, who checks its output, its controls and the four gaps.
The Model Risk CommitteeThe agenda, the decisions asked of it and its members. The pack closes the Friday before.
The bank’s policyWhat counts as a model, tier lines, validation timing, control limits and rules, finding due dates and approvers.
Governance

Validation sign-off and the audit trail in model risk management software

Each decision cites its sourceEach scope decision cites the SR 26-2 passage, and each finding the document page or test run behind it. A validation report carries its citations: 31 in the sample CRE model’s draft.
Agents propose, a person decidesBorderline spreadsheets always go to the head of model risk management. Scope, tiers and the validation plan reach the committee only after that approval.
Two signatures on each validationThe head of model risk management signs as approver, with the meaning of the signature recorded, and the lead validator co-signs. Conditional use and extensions for Tier 1 models go to the Model Risk Committee.
Nothing leaves without a reasonNo entry is deleted. One that leaves the inventory keeps its owner, history and reason, so internal audit can open it and test the decision.
Findings close on a validator’s reviewThe agents chase owners and draft reminders and extension requests; only a validator closes a finding, on the remediation evidence. A signed report cannot be changed. A new finding is raised instead.
Each step on the audit trailEach agent run and human decision is recorded with who, what and when. Policy settings are versioned, and the solution’s own agents sit in the AI tool register like any other agentic tool.
Configuration

Tier lines, control limits and finding due dates from the model risk policy

The agents apply the bank’s own policy. Changes are versioned and on the audit trail.

SettingDefaultChoose from
Spreadsheets with forecasting logicA person decidesA person decides · Keep as models · Treat as tools
Tier 1 exposure line$1 billion$500 million · $1 billion · $2 billion
Tier 1 validationEvery 12 monthsEvery 12 · 18 · 24 months
Control limits±3σ from the first 12 months in production±3σ · ±2.5σ · ±2σ
Control rules8 on one side · 6 moving8 on one side · 6 moving · 9 on one side · 6 moving · Limits only
When a limit is crossed (Tier 1 and 2)Open a revalidation taskOpen a revalidation task · Alert the owner only
Calibration testAmber 95% · red 99%Amber 95% · red 99% · Amber 90% · red 95% · Red 99% only
Finding due dates, from the committee dateHigh 90 · Medium 180 · Low 365 daysHigh 90 · Medium 180 · Low 365 days · High 60 · Medium 120 · Low 270 days
Connections

Model inventory, production outcomes and validation files

Your model inventoryevery entry, owner, tier and validation date
Production scores and outcomesmonthly defaults, scores and overrides from the loan books
Development and vendor documentationdevelopment documents, user guides, test logs
Prior validations and findingspast reports, open findings, monitoring reports
Change logfeed, data and code changes drawn on each chart
AI tool use casesgenerative and agentic tools, their owners and controls
What it changes

Drift caught between validations, and a validation plan the team can staff

Each figure carries its label: a target the solution is built to, an estimate, a typical published result, or a proven one.

target
Monthly
drift checks on production data, so a breach opens revalidation that month
Annual validationup to 12 months
Control charts1 month
estimated
131validations
due over 12 months in the sample case, down from 214, once timing follows materiality
Old plan · 34,600 hBy materiality · 22,320 h
target
100%
of findings cited to the page or test run that proves them
4 of 4 findings
on every validation

“demo” = seen in the working solution, on its sample bank data · “target” = the design goal, measured in the live solution · “estimated” = our estimate of where a validation’s days go. Regulation: Federal Reserve, OCC and FDIC, SR 26-2 Supervisory Guidance on Model Risk Management (17 April 2026; replaces SR 11-7 and SR 21-8; generative and agentic AI outside its scope). Names on the screens are sample data.

Questions

Model risk management software questions on SR 26-2, drift and AI tools

What is model risk management software?

Model risk management software keeps a bank’s model inventory, tiers each model by materiality, plans and records independent validation, monitors models in production and tracks findings to closure. Model Risk Management adds agents that score each model monthly on production data, re-scope the inventory, read the evidence and draft the validation report, while people decide and sign.

How does it re-scope our model inventory to SR 26-2?

The inventory scoper tests each entry against the SR 26-2 definition of a model and cites the passage; the materiality tierer proposes a tier from exposure, purpose and inherent risk. Simple calculations move to a tools register, generative and agentic AI to an AI tool register, each with its reason, and borderline spreadsheets go to the head of model risk. The committee ratifies the result.

How does it catch model drift between validations?

Each model gets a control chart for accuracy, stability and overrides, with a center and ±3σ limits set from its first 12 months in production. The outcomes analyst scores it each month on production data; a control rule turns the model amber before a limit is crossed, and a breach on a Tier 1 or Tier 2 model opens a revalidation task with its evidence.

Does the AI write and sign our validation reports?

It drafts them, with citations, a proposed severity for each finding and a proposed rating, and it never signs. The validator can change any severity; the head of model risk management signs as approver and the lead validator co-signs.

How does it govern generative and agentic AI tools?

SR 26-2 leaves them out of scope, so each tool goes into an AI tool register with its owner, what it does, whether it drafts, informs or acts, who checks its output, its controls and its review date. The AI tool registrar flags control gaps; it does not approve a tool for use.

Can internal audit see why a model left the inventory?

Yes. Nothing is deleted. Each entry keeps its owner, history and the reason for its scope and tier, with the passage it rests on, and internal audit reads it with read-only access.

Can we set our own tiers, cadence and thresholds?

Yes. Tier lines, validation timing by tier, control limits and rules, PSI, AUC and calibration thresholds, finding due dates and approvers are settings, versioned and on the audit trail.

Which of our records does the build start from?

Your model risk policy, your inventory, your validation report template and a sample of past validations. The Agentic Solution Engine builds and deploys it from those requirements, and it goes live once every quality gate has passed. We will walk you through it on your own models first.

Score last month’s production data
with model risk management.

We’ll run Model Risk Management on a sample of your own inventory and monitoring data.