What features could Riskified add to its Policy Protect solution to better detect and prevent return abuse?

Introduction

To improve Riskified's Policy Protect solution for better detection and prevention of return abuse, we need to analyze the current product, understand user pain points, and develop innovative features. I'll outline a strategic approach to enhance this critical component of Riskified's fraud prevention suite.

Step 1

Clarifying Questions (5 mins)

• Looking at Riskified's position in the market, I'm thinking about the scale of data they process. Could you share more about the volume of transactions Policy Protect handles daily and the types of merchants using it?

Why it matters: Determines the scale of our solution and potential resource constraints Expected answer: Processes millions of transactions daily for large e-commerce retailers Impact on approach: Would focus on scalable, automated solutions with minimal manual review

• Considering the evolving nature of return fraud, I'm curious about the current accuracy rates of Policy Protect. What are the false positive and false negative rates for detecting return abuse?

Why it matters: Helps identify areas for improvement in the detection algorithm Expected answer: False positive rate around 5%, false negative rate around 2% Impact on approach: Would prioritize reducing false positives to improve merchant satisfaction

• Given the potential for new fraud patterns, how frequently is Policy Protect updated with new rules or machine learning models?

Why it matters: Indicates the agility of the current system and potential for rapid improvements Expected answer: Major updates quarterly, with minor tweaks monthly Impact on approach: Would consider a more dynamic, self-learning system for quicker adaptation

• Thinking about integration with merchant systems, what level of customization does Policy Protect currently offer for different retail environments?

Why it matters: Determines if we need to focus on flexibility and integration features Expected answer: Basic customization options available, but limited to predefined parameters Impact on approach: Would explore more advanced customization and API capabilities