The Challenge
A rapidly growing fintech lender was making credit decisions using manual processes and spreadsheet-based scoring models. Decision times averaged 4 hours, analyst capacity was a bottleneck on growth, and model performance was declining as the customer base expanded beyond the original demographic.
Our Approach
We designed and built a modern data platform on AWS using dbt, Airflow, and Snowflake as the core stack. On top of this foundation, we developed a gradient boosting credit scoring model trained on 18 months of historical decision and outcome data. The model was deployed via a real-time API integrated directly into the loan origination system.
We also built a comprehensive model monitoring and governance framework — essential for a regulated environment — including drift detection, fairness monitoring, and a model card aligned to FCA explainability expectations.
The Results
- Credit decision time reduced from 4 hours to under 90 seconds
- Model AUC improved from 0.71 to 0.84 vs. the previous scorecard
- Analyst capacity freed to focus on complex edge cases (estimated value: 3 additional analysts)
- Model passed FCA governance review on first submission