🔹 How the Model Validation Process at Banks Can Drive Business Impact??
Link: https://www.linkedin.com/embed/feed/update/urn:li:share:7291489903228436480
Model validation isn’t just compliance—it directly impacts revenue, risk, and decision-making. As a risk model validator, I usually experience the confused eyes from stackholders when trying to explain them technical terms related to statistics, data mining, or even coding. In order to have common language with business units and increase the impact of validation function, it’s more convenient to instead of focusing only on PSI, AUC, or KS, we must connect its insights to financial impact.
🔹 1. Convert Metrics into Business Language
– Instead of: “PSI increased by 0.2, indicating data drift.”
– Say this: “If not fixed, misclassification may cost $20M in lost revenue per year.”
Why? Business leaders care about profits and risk exposure, not technical numbers.
🔹 2. Align Validation with Revenue Growth
– Before: Validation was just a model check.
– Now: It optimizes approvals, reduces risk, and increases revenue.
Example: Adjusting thresholds helped approve 5,000 more customers monthly, unlocking $100M in additional loans.
🔹 3. Use Insights to Guide Strategy
– Don’t just report issues—propose solutions.
Example: Instead of just saying “AUC dropped from 0.85 to 0.78,” explain:
“The model’s predictive power fell 8%, which could increase credit losses by $30M. We recommend retraining with updated macroeconomic variables.”
🔹 4. Make Validation a Strategic Function
– Shift from: “Validation ensures compliance.”
– To: “Validation drives revenue and risk strategy.”
📌 Key Actions:
✔ Present validation impact in financial terms.
✔ Integrate insights into business & executive reviews.
✔ Suggest alternative strategies to improve risk-reward balance.
💡 Key Takeaway:
When validation links to business impact, it shifts from a technical task to a strategic growth driver and make the validation function become more valuable to banks’ goals. 🚀
🔽 How does your team integrate validation with business?
#ModelValidation #RiskManagement #BusinessImpact #creditscoringmodels #statisticalmodels #machinelearning
Jayan Madusanka
Senior Assistant Manager @ Commercial Bank of Ceylon PLC | Banker | Business Analytics | AI & Data Science | Analytics Engineering | Credit Risk Modeling
Denise Nguyen Agreed 💯. The business users / underwriters don’t understand the technical jargons. They are merely worried about whether these models outperform the human lending decision, what the improvement for TAT, how accurate the decision made by the model, etc. Meanwhile the Risk officers are concerned about the portfolio health and impact of the existing credit policies. Mostly the underlying dataset has class imbalance due to the prudent lending practises. What are your thoughts to overcome that?
Reply
Denise Nguyen
Author
Jayan Madusanka changing bad flag to smaller dpd, using other criterion to define bad such as re-structured loan, written-off loan. I used to train models with 3.5% by LogReg and it still had gini over 50%.
