Skip to main content
Skip to footer

Streamlining ML-based Model Management Framework: A BFS Industry Perspective

 

Ushasi Sengupta

Senior Research Analyst, Corporate Marketing Research, TCS

Sanjukta Dhar

Lead, CRO Strategic Initiative-Canada, BFSI, TCS

Ensuring interpretability and reduced time-to-market for new-age ML models

The adoption of artificial intelligence (AI) and machine learning (ML) has gone up manifold across the value chain of financial risk management models with the aim of driving efficiency and diverse data-driven analytics. With the increasing use of AI in risk management, institutions need to reimagine the traditional model risk management framework and review critical aspects in order to reduce time-to-market for new-age ML models and ensure optimal performance.

The three critical components of ML model risk management are:

Data validation: Discovery, preprocessing, feature engineering, and drift monitoring for the model input data

Outcome validation: Cross-validation with various sets of data to validate outcome and assess performance deviation

Model oversight and control: Adopt a resilient model oversight framework by defining stringent model design development policies, periodic reviews, and materiality scoring systems