Card-Not-Present Fraud Detection for a Market Dominated by Online Spending
Company: Visa
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
You are a product manager at Visa. In one geography, most consumers make their card purchases online. How would you detect fraudulent transactions there?
```hint Start from what online changes
When the card is not physically present, some defenses that work at a store terminal no longer apply. Think about which signals an online payment still carries, and who in the payment chain sees each of them.
```
### Clarifying Questions
- Should fraud be caught in real time while the transaction is being authorized, found after the fact, or both?
- Whose losses and experience matter most here: card issuers, merchants or cardholders?
- Is the main concern stolen card details used online, or also account takeover at merchants, cardholders disputing purchases they really made, and fraudulent merchants?
- Which data can be used in this geography, such as device and browser details passed during online authentication, and do local privacy or data-residency rules limit it?
### What a Strong Answer Covers
- Framing detection as a real-time decision that weighs the cost of missed fraud against the cost of declining good customers.
- The signals available on online transactions, and where in the payment flow each one can be used.
- A layered approach (prevention, rules, a machine-learned risk score, step-up authentication), including how labels are obtained and the model is trained.
- Evaluation and monitoring with metrics such as fraud losses, false declines and approval rate, plus drift detection.
- Considerations specific to this geography, such as cross-border shopping, local regulation and limited fraud history.
### Follow-up Questions
- Fraud losses fall after launch, but merchants report more good customers being declined. How would you find the right threshold?
- Chargeback labels arrive weeks after the transaction. How does that delay affect training and monitoring?
- A new type of online merchant grows quickly in the region and has no fraud history. How would you score its transactions?
Overview: A Visa product manager question on detecting fraudulent card transactions in a market where most spending happens online. It tests understanding of card-not-present fraud, the payment flow and signals involved, layered detection with rules and machine learning, and the balance between fraud losses and false declines.