For Lendable (London)'s credit scoring model, should we focus on improving accuracy or reducing bias, given limited resources for model development?

Introduction

For Lendable's credit scoring model, we're facing a critical trade-off between improving accuracy and reducing bias, given our limited resources for model development. This decision will significantly impact our lending practices, customer base, and overall business performance. I'll analyze this trade-off by examining the product context, key metrics, experimental approaches, and potential outcomes to provide a strategic recommendation.

Step 1

Clarifying Questions (3 minutes)

• Based on the fintech industry trends, I'm thinking Lendable might be facing increased regulatory scrutiny. Could you share any recent regulatory changes or upcoming requirements that might impact our decision?

Why it matters: Helps prioritize compliance and risk management in our approach Expected answer: New regulations requiring stricter fairness standards in lending Impact on approach: Would shift focus towards bias reduction and model transparency

• Considering Lendable's business model, I assume improving accuracy could lead to better loan performance. What's our current default rate, and how does it compare to industry benchmarks?

Why it matters: Helps quantify the potential impact of improved accuracy on our bottom line Expected answer: Default rate slightly above industry average Impact on approach: Would strengthen the case for prioritizing accuracy improvements

• Looking at our user segments, I'm curious about the diversity of our applicant pool. Can you provide insights into the demographic breakdown of our current customer base?

Why it matters: Helps assess the potential impact and urgency of addressing bias in our model Expected answer: Customer base skews towards certain demographics Impact on approach: Would emphasize the importance of bias reduction for long-term growth

• Regarding our technical capabilities, I'm wondering about our current model architecture. Are we using traditional statistical methods or more advanced machine learning techniques?

Why it matters: Influences the feasibility and approach to improving accuracy or reducing bias Expected answer: Mix of traditional and ML models with room for improvement Impact on approach: Would inform the technical strategy for model enhancements

• Considering resource constraints, what's our current data science team capacity and budget allocation for model development?

Why it matters: Helps determine the scope and timeline of potential improvements Expected answer: Limited team capacity with moderate budget for external resources Impact on approach: Would necessitate prioritization and phased implementation of improvements