Executive Summary
As AI Governance Lead, I ran a high-risk AI governance assessment of Meridian Financial Services' pilot for an automated small business loan underwriting system — scoping the risk classification, running the review against three major frameworks, and producing the evidence package a governance committee would need to make a production decision.
The system uses a third-party AI model, CrediSure Credit Decision Engine v2.3, to evaluate small business loan applications and produce one of three outcomes: auto-approve, auto-deny, or route to manual review.
Because approximately 94% of applications are processed automatically and the system affects access to credit, I classified this as a high-risk AI use case requiring governance review before production deployment.
Skills Demonstrated
Frameworks Applied
| EU AI Act | NIST AI Risk Management Framework | ISO/IEC 42001 |
|---|---|---|
| High-risk AI classification | Governance roles and responsibilities | AI management system scope |
| Fundamental Rights Impact Assessment | System context and intended use | Organizational roles and responsibilities |
| Human oversight | Risk measurement and testing evidence | AI risk assessment |
| Logging | Risk prioritization and treatment | AI risk treatment |
| Explanation to affected persons | Human-AI oversight | Management review |
| Appeal and contestability | AI system lifecycle management | |
| Deployment conditions | Third-party AI risk management |
Central Governance Question
My answer: Not for unrestricted production deployment. The system may proceed only with conditions because the current governance evidence does not fully support safe, fair, explainable, accountable, and well-monitored deployment.
Portfolio Artifacts
- AI System Profile and Intake Record Documents system purpose, users, affected groups, data, vendor, and risk classification.
- AI Risk Register and Mitigation Summary Documents key AI risks, severity, controls, residual risk, and recommendations.
- Human Oversight and Appeal Procedure Defines human review triggers, override authority, escalation, and applicant appeal process.
- Third-Party Vendor and Model Review Evaluates CrediSure AI vendor risk, model limitations, and required evidence.
- Production Readiness Decision Memo Provides final executive recommendation.
- Final VerifyWise Portfolio Report Consolidated portfolio report generated from VerifyWise.
Practical AI Governance Skill Proof
1. Use Case Registration
2. Model Inventory
3. Dataset Record
4. AI Risk Register
Key Risks Identified
| Risk | Why It Matters |
|---|---|
| Discriminatory lending outcomes | The system may unfairly approve or deny applicants based on biased data, proxy variables, or historical lending patterns. |
| Proxy bias through credit and business variables | Variables such as geography, business age, industry, credit history, thin credit files, or cash-flow volatility may create unfair outcomes. |
| Weak explainability and incomplete reason codes | Meridian may be unable to explain automated denials, support appeals, or demonstrate compliance during review. |
| Accountability gaps in third-party AI deployment | Meridian remains responsible for deployment even if the vendor controls key model details. |
| Inaccurate automated denials | Qualified applicants may be incorrectly denied credit due to model error, incomplete data, or overly strict thresholds. |
| Lack of meaningful human oversight | Routing only 6% of applications to human review may be insufficient for a high-risk credit decision system. |
5. Vendor Record
6. Framework Assessments
7. Fundamental Rights Impact Assessment (FRIA) Summary
8. Final Report
Oyinadesola Edu_AI_Governance_Portfolio_Report.pdf
Final Recommendation
Meridian should not approve unrestricted production deployment at this time. The system may move forward only if the following conditions are completed:
- Complete fairness testing and disparate impact analysis
- Complete proxy-bias review
- Validate denial reason codes
- Define mandatory human review triggers
- Establish appeal and reconsideration procedures
- Complete vendor documentation review
- Complete security and privacy review
- Define model and outcome monitoring thresholds
- Implement decision-level audit logging
- Establish incident escalation procedures
- Obtain governance committee approval
Portfolio Conclusion
This project demonstrates my ability to evaluate a high-risk AI system from an AI governance, risk, and compliance perspective. The Meridian Automated Loan Underwriting System offers business benefits, including faster decisions, improved consistency, and operational efficiency. However, because it affects access to credit and automates most decisions, it requires strong governance before deployment.
As AI Governance Lead, my assessment found that the system should not receive unrestricted production approval until Meridian completes required fairness, explainability, human oversight, vendor, monitoring, privacy, and governance controls.