What is a Decision Dossier?Return to Workspace

Representative demonstration using fictionalized institutional data. Not investment, credit, legal, or regulatory advice.

Sample Decision Dossier

AI-Assisted Credit Decisioning for Retail Lending

Decision ID: DEC-2026-0417

BOARD PUBLICATION BLOCKED

Institution: Fictional Tier-One Regional Bank
Sponsor: Chief Risk Officer
Materiality: $38,700,000
Prepared: August 28, 2026
Version: v1.3
Status: BOARD PUBLICATION BLOCKED

Executive Decision Brief

Decision Being Considered

Whether to approve full production rollout of an AI-assisted credit decisioning model across the retail lending portfolio, replacing the existing rules-based scorecard for unsecured consumer credit originations.

Strategic Rationale

The model is projected to reduce expected credit losses by improving risk segmentation at the margin, while increasing approval throughput for a segment of near-prime applicants the current scorecard under-serves. The institution has already committed engineering and vendor spend to build the model; the decision under review is whether to authorize its use in live credit decisions.

Recommended Disposition

HOLD PENDING EVIDENCE REFRESH

Principal Findings

1.The model’s reason-code coverage was last validated against a feature set that predates the July 2026 feature-engineering change, and has not been retested against the current production feature set.
2.Disparate-impact testing on the pre-July feature set showed no statistically significant adverse impact by protected class, but this finding cannot be extended to the current production model without retesting.
3.Expected loss reduction and NPV figures in the financial model remain directionally sound across all four scenarios, including the stress case, independent of the reason-code question.

Conditions Required Before Approval

Reason-code coverage retested against the current production feature set, with results reviewed by Model Risk.
Disparate-impact testing rerun against the current production feature set and reviewed by Compliance.
Data Governance sign-off on the July feature-set change log and its downstream effect on adverse action notice generation.

Decision Mandate

Scope

Applies to all unsecured consumer credit origination decisions made through the retail lending platform, including new-account approvals, line assignment, and initial pricing tier. Excludes secured lending, commercial credit, and existing-account line management, which are governed under separate mandates.

Decision Rights

The Chief Risk Officer holds decision rights to approve production rollout, subject to Model Risk Management and Compliance sign-off on all Constitutional articles below. The Board Risk Committee holds override authority and must be briefed before any production rollout affecting more than 15% of origination volume.

Accountable Executive

Chief Risk Officer

Approval Authority

Board Risk Committee (final publication authority)

Evaluation Period

February 2026 – August 2026, with quarterly re-evaluation post-launch

Constitution Articles Applied

I. MandateII. Permitted UsesIII. Prohibited UsesIV. Hard LimitsV. Escalation TriggersVI. Intervention Authority

Classified Evidence Ledger

Evidence ID	Claim	Classification	Source	Owner	Freshness	Status	Contradiction
E-01	Backtested Gini coefficient of 0.61 on the current model versus 0.54 for the incumbent scorecard, on a 24-month holdout sample.	Evidence	Model Validation Report v4.2	Model Risk Management	Current — revalidated July 2026	Verified	No
E-02	Projected expected-loss reduction of 40–55 basis points annually across the affected origination segment.	Inference	Financial Model v3, Base Case	Finance — Credit Risk Analytics	Current	Verified	No
E-03	Approval-rate increase of 6.2 percentage points for near-prime applicants (FICO 620–679) without a corresponding increase in 90-day delinquency.	Evidence	Champion/Challenger Pilot, Q1–Q2 2026	Retail Lending Analytics	Current — pilot closed June 2026	Verified	No
E-04	No statistically significant disparate impact by race, ethnicity, or sex was detected in the pre-July feature set at the 5% significance threshold.	Evidence	Fair Lending Disparate Impact Analysis, May 2026	Compliance — Fair Lending	Stale — predates July feature-set change	Contested	Yes
E-05	Vendor-supplied model explainability layer generates SHAP-based reason codes mapped to the four adverse-action reason categories required by Regulation B.	Evidence	Vendor Technical Specification, Section 7	Data Governance	Current	Verified	No
E-06	The July 2026 feature-engineering change added three derived income-stability features and removed one deprecated bureau-trade feature.	Evidence	Feature-Set Change Log, July 2026	Data Governance	Current	Verified	No
E-07	Model latency at production scale (P99) is 340ms, within the 500ms SLA required by the origination platform.	Evidence	Production Load Test Report, August 2026	Engineering — Platform Reliability	Current	Verified	No
E-08	Reason-code coverage — the percentage of declined applications for which the model’s top adverse-action reason code correctly matches a human-reviewable factor — was measured at 97.3%.	Evidence	Reason-Code Coverage Test, March 2026	Model Risk Management	STALE — predates July feature-set change, not retested	Blocking	Yes
E-09	Legal has confirmed the vendor contract includes indemnification for model performance claims but excludes indemnification for regulatory fines arising from the bank’s use of model output.	Evidence	Vendor Contract Review Memo, June 2026	Legal	Current	Verified	No
E-10	If reason-code coverage on the current production feature set falls below 95%, the model cannot generate compliant adverse-action notices and rollout must not proceed.	Judgment	Model Risk Management, escalation memo	Model Risk Management	Current	Verified	No
E-11	Portfolio-level credit losses are assumed to remain within the bank’s current capital planning assumptions for the retail segment through the 2027 planning horizon.	Assumption	Capital Planning Working Assumptions, 2026	Finance — Capital Planning	Current	Assumed	No
E-12	Customer complaint volume related to adverse-action notices is expected to remain flat or decline slightly, based on the pilot’s complaint data.	Inference	Customer Experience Pilot Review, June 2026	Customer Experience	Current	Verified	No

Financial Model

All figures below are representative and computed for this fictionalized case. They are not a forecast or guarantee of any real-world outcome.

Five-Year TCO

$14,200,000

NPV

$61,400,000

IRR

47%

Payback Period

1.8 yrs

Expected Loss Reduction

40–55 bps annually

Implementation Cost

$9,600,000

Compliance & Remediation Reserve

$2,100,000

Scenario Analysis

Base

$61,400,000

NPV: $61,400,000
Cost: $14,200,000
Expected Benefit: Expected loss reduction of ~47 bps; approval-rate gain of 6.2pp in near-prime segment.
Risk Exposure: Moderate — dependent on reason-code retest passing at current thresholds.
Decision Implication: Proceeds once evidence refresh conditions are satisfied.

Upside

$84,900,000

NPV: $84,900,000
Cost: $13,100,000
Expected Benefit: Loss reduction reaches 60+ bps as near-prime segment volume grows faster than forecast.
Risk Exposure: Low-moderate — upside case still requires the same regulatory conditions to be met.
Decision Implication: Does not change the evidence-refresh requirement; publication remains blocked until E-08 is resolved.

Downside

$38,700,000

NPV: $38,700,000
Cost: $16,800,000
Expected Benefit: Loss reduction compresses to ~25 bps if near-prime delinquency normalizes upward.
Risk Exposure: Moderate-high — narrower margin for error on fair-lending and reason-code performance.
Decision Implication: Still NPV-positive, but the evidence-refresh conditions become more, not less, important.

Stress

$11,300,000

NPV: $11,300,000
Cost: $19,500,000
Expected Benefit: Loss reduction falls to ~12 bps under a retail-credit stress cycle with rising unemployment.
Risk Exposure: High — model behavior under stress has not been independently validated outside the vendor’s own stress test.
Decision Implication: Remains NPV-positive but materiality of an unresolved reason-code gap increases sharply in a stress cycle.

Twelve-Lens Assessment

1. Financial Impact

Clear

NPV-positive across all four scenarios, including stress. Payback under two years in the base case.

Mitigation: None required.

Accountable Owner: Finance — Credit Risk Analytics

2. Technical Debt

Watch

The July feature-set change was implemented without a corresponding update to the model documentation package.

Mitigation: Documentation package to be updated before production sign-off.

Accountable Owner: Engineering — Model Platform

3. Security Exposure

Clear

Model runs within the existing decisioning platform’s security boundary; no new external attack surface introduced.

Mitigation: None required.

Accountable Owner: InfoSec

4. Compliance

Blocking

Reason-code coverage (E-08) has not been retested against the current production feature set. Adverse-action notice generation cannot be certified compliant until retested.

Mitigation: Retest reason-code coverage on current feature set; Compliance sign-off required before publication.

Accountable Owner: Chief Compliance Officer

5. Operational Resilience

Clear

Fallback to the incumbent rules-based scorecard is fully built and tested for platform outages.

Mitigation: None required.

Accountable Owner: Engineering — Platform Reliability

6. Customer Impact

Watch

Near-prime applicants gain approval access, but any applicant declined under a reason-code gap receives a notice that may not reflect the true decision driver.

Mitigation: Resolved once E-08 is retested and coverage confirmed above the 95% threshold.

Accountable Owner: Customer Experience

7. Talent

Clear

Existing Model Risk and Retail Analytics staff have the capacity and skill set to operate and monitor the model post-launch.

Mitigation: None required.

Accountable Owner: Human Resources — Risk & Analytics

8. Competitive Position

Watch

Two direct competitors have already deployed comparable near-prime underwriting models; delay carries a quantifiable share-of-originations cost.

Mitigation: Time-box the evidence-refresh work to a defined window rather than an open-ended pause.

Accountable Owner: Retail Lending Executive

9. Vendor Lock-in

Watch

The explainability layer generating reason codes is proprietary to the vendor and not portable to an alternate model platform without re-engineering.

Mitigation: Contract renewal terms to include a data- and logic-portability clause at next renewal.

Accountable Owner: Procurement — Vendor Management

10. Scalability Ceiling

Clear

Load testing confirms headroom to 3x current origination volume within existing infrastructure.

Mitigation: None required.

Accountable Owner: Engineering — Platform Reliability

11. Reversibility

Clear

Rollback to the incumbent scorecard is a same-day operation with no customer-facing disruption.

Mitigation: None required.

Accountable Owner: Engineering — Platform Reliability

12. Reputational Risk

Blocking

Publishing an approved dossier while a fair-lending-relevant test (E-08) is stale would be indefensible if surfaced by an examiner or in litigation discovery.

Mitigation: No publication until E-08 is retested; this is the specific finding that blocks board publication.

Accountable Owner: Chief Risk Officer

Challenge Room & Dissent Record

Model Risk OfficerBlocking

Represented by Model Risk Management

The reason-code coverage figure in the dossier (E-08) is being carried forward from a test run against a feature set that no longer exists in production. Citing it as current evidence is a validation error, not a technicality.

Chief Compliance OfficerBlocking

Represented by Compliance

Compliance cannot certify Regulation B adverse-action notice compliance for the current production model on the basis of a pre-July test. This is a hard blocker, not a residual risk to be accepted.

Retail Lending ExecutiveAccepted as Residual Risk

Represented by Retail Lending

A prolonged, open-ended hold risks ceding near-prime origination share to competitors who have already launched comparable models. Requests a time-boxed retest window rather than an indefinite pause.

Data Governance LeadResolved

Represented by Data Governance

The July feature-set change was logged and version-controlled correctly (E-06), but the change-management process did not trigger an automatic re-validation requirement for downstream compliance tests. That process gap should be fixed independent of this decision.

Decision Constitution Status

I. MandateSatisfied

Decision scope and accountable executive must be named and current.

Evidence: Mandate section, this dossier

II. Permitted UsesSatisfied

Model use limited to unsecured retail credit origination decisions as scoped.

Evidence: Mandate scope statement

III. Prohibited UsesSatisfied

Model output must not be used for existing-account line management or collections prioritization without separate Constitutional review.

Evidence: Mandate scope statement

IV. Hard LimitsUnsatisfied

Reason-code coverage on the production feature set must exceed 95% before go-live.

Evidence: E-08 (stale — not retested against current feature set)

Exception Owner: Chief Risk Officer

V. Escalation TriggersEscalation Triggered

Any unresolved blocking dissent from Model Risk or Compliance automatically escalates to the Board Risk Committee.

Evidence: Challenge Room record — two blocking objections outstanding

Exception Owner: Board Risk Committee

VI. Intervention AuthoritySatisfied

The Board Risk Committee retains authority to pause or reverse production use at any time post-launch.

Evidence: Mandate, decision rights statement

Approval Chain

Role	Name / Function	Status
Prepared by	Decision Workspace — Retail Credit Risk	Reviewed
Evidence owners	Model Risk Management, Data Governance, Finance	Reviewed
Model Risk	Model Risk Management	Blocked
Compliance	Chief Compliance Officer	Blocked
Finance	Chief Financial Officer	Conditionally Approved
Executive sponsor	Chief Risk Officer	Conditionally Approved
Board committee	Board Risk Committee	Pending

Statuses reflect workspace record only. No signatures are fabricated or implied.

Final Decision

BOARD PUBLICATION BLOCKED

Publication to the board is blocked until reason-code coverage (E-08) is retested using the current production feature set and the results are reviewed and signed off by both Model Risk Management and Compliance. This is a evidence-freshness failure, not a finding of model unsoundness — the underlying financial and portfolio case remains strong across all four scenarios. The dossier will be resubmitted for board publication once the retest is complete and both Constitutional Article IV and the two outstanding blocking objections in the Challenge Room are resolved.

Required Actions

Action	Owner	Deadline	Dependency	Completion Evidence
Retest reason-code coverage against current production feature set	Model Risk Management	September 18, 2026	None — can begin immediately	Updated E-08 test report showing coverage ≥ 95%
Rerun disparate-impact analysis on current feature set	Compliance — Fair Lending	September 25, 2026	Requires reason-code retest output	Updated fair-lending disparate impact memo
Update model documentation package to reflect July feature-set change	Engineering — Model Platform	September 30, 2026	None	Signed-off documentation package v5.0
Amend change-management process to trigger automatic compliance re-validation on feature-set changes	Data Governance Lead	October 15, 2026	None	Updated change-management SOP
Resubmit dossier to Board Risk Committee for publication	Chief Risk Officer	October 22, 2026	All prior actions complete; Model Risk and Compliance sign-off received	Board Risk Committee meeting minutes

Audit Trail

Timestamped, representative activity log. Every entry below corresponds to a state change described elsewhere in this dossier.

2026-02-11 09:14
Decision framed and scoped; Mandate drafted by Chief Risk Officer.
2026-03-04 14:02
Evidence E-08 (reason-code coverage) added by Model Risk Management.
2026-05-19 11:30
Evidence E-04 (disparate impact) added by Compliance — Fair Lending.
2026-06-30 16:45
Assumption logged: portfolio losses remain within 2027 capital planning envelope.
2026-07-08 10:12
Feature-set change (E-06) deployed to production by Data Governance.
2026-08-02 13:20
Scenario analysis recalculated across Base, Upside, Downside, and Stress cases.
2026-08-14 09:55
Dissent recorded: Model Risk Officer flags E-08 as stale relative to the July feature-set change.
2026-08-15 11:08
Dissent recorded: Chief Compliance Officer blocks certification pending retest.
2026-08-20 15:41
Constitutional Article IV marked unsatisfied; Article V escalation trigger fired to Board Risk Committee.
2026-08-28 08:00
Board publication blocked. Dossier version 1.3 finalized pending evidence refresh.

Use Print / Save as PDF for a print-friendly, single-column version of this page.

Request a Decision Dossier Review Return to Workspace