Credit Risk Model — Requirements & Compliance

The Problem

Legacy rule-based credit model had 65% approval rate but a high default rate. Static rules rejected creditworthy customers and couldn't capture complex behavioral patterns. Regulators required full explainability for every credit decision — a constraint that had to be built into the model design from the start.

My Contribution

I led requirements elicitation with credit adjudication, risk, and compliance stakeholders to define the model inputs, approval thresholds, and regulatory explainability requirements. I mapped the existing rule set to the feature engineering requirements, documented data lineage for all source attributes used in the model, and worked with compliance to ensure the explainability approach satisfied regulatory audit standards before the model was built. I ran UAT with the credit adjudication team using historical case samples and authored the operational runbook for ongoing model monitoring and governance.

The Solution

Requirements-led model development: regulatory explainability specifications defined upfront, data lineage documentation for all 200+ features, compliance-approved threshold definition, structured UAT with historical cases, and an operational governance runbook for post-launch monitoring.

Results

• Approval rate improved: 65% → 88% (+23%)
• Default rate reduced: 8% → 6.8% (-15%)
• 100ms prediction latency
• 92% AUC on holdout test set
• $15M+ added annual revenue

Tech Stack