PayPal·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say

May 2026

Summary

PayPal payments-risk case for a data scientist role. The whole thing was one big scenario question about real-time fraud decisioning, which I was not fully prepared for in terms of how deep they wanted to go on the business trade-offs.

Questions Asked (1)

You're on the payments-risk team and two new transactions come in with attributes like amount, issuing country, device type, historical fraud rate, and card age. Walk through your real-time decision process for each transaction, name at least three risk strategies you'd deploy right away, and explain the trade-offs between false positives and chargebacks.

Technical Trade-offsProduct Analytics & MetricsA/B Testing & Experimentation

Author's notes

I started with the obvious stuff, velocity checks and rule-based thresholds, but then kind of froze when they pushed on how I'd actually set those thresholds.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around a real-time decision pipeline: first assess each transaction's risk using the given attributes, then apply specific risk strategies, and finally discuss the trade-offs between false positives and chargebacks. Emphasize that the goal is to balance fraud prevention with customer experience, using data-driven thresholds and continuous monitoring.

Pro tip: Quantify the trade-offs: for example, mention that a 1% increase in false positives can lead to significant customer friction and lost revenue, while a 1% increase in chargebacks can result in fines and higher processing fees. Show that you understand the business impact beyond just model accuracy.

1. Assess Transaction Risk

For each transaction, evaluate the provided attributes (amount, issuing country, device type, historical fraud rate, card age) to form an initial risk score. Consider how each attribute contributes to risk, e.g., high amount, high-risk country, new device, high historical fraud rate, and new card age increase risk.

2. Apply Risk Strategies

Deploy at least three risk strategies in real-time: (1) rule-based filters (e.g., block transactions from high-risk countries or above a certain amount), (2) machine learning model scoring (e.g., a gradient boosting model that outputs a fraud probability), and (3) velocity checks (e.g., number of transactions per card/device in a short time window). Optionally, mention additional strategies like device fingerprinting or behavioral analytics.

3. Decide on Action

Based on the combined risk score and strategy outputs, decide whether to approve, decline, or flag for manual review. Set thresholds that balance risk and customer experience, and consider using a step-up authentication for medium-risk transactions.

4. Explain Trade-offs

Discuss the trade-off between false positives (legitimate transactions declined) and chargebacks (fraudulent transactions approved). False positives lead to customer dissatisfaction and lost revenue, while chargebacks incur fees, penalties, and potential loss of merchant privileges. Highlight the need to optimize thresholds using cost matrices and A/B testing.

5. Monitor and Iterate

Emphasize the importance of continuous monitoring, feedback loops, and model retraining to adapt to evolving fraud patterns. Mention metrics like precision, recall, F1, and business KPIs like fraud rate and false positive rate.

Key Points to Mention

Real-time scoring using a machine learning model (e.g., logistic regression, random forest, or gradient boosting) that incorporates all available features.
Rule-based strategies such as blocking high-risk countries, high amounts, or mismatched geolocation.
Velocity checks to detect rapid successive transactions from the same card, device, or IP.
Cost-sensitive threshold optimization to balance false positives and chargebacks, using a cost matrix that accounts for chargeback fees, customer lifetime value, and operational costs.
A/B testing or champion-challenger frameworks to evaluate new risk strategies in production.
Continuous model monitoring and retraining to combat concept drift and adversarial fraud patterns.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.