How Ad Platforms Use Audience, Context and Performance Data to Choose Which Ad to Serve
Company: Rakuten
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
As a product manager working on advertising products at Rakuten, explain how data is used to decide which ad is served when an ad slot needs to be filled. Cover what data the decision draws on, where in the serving process each kind of data is used, and how the results of ads already served shape later decisions.
```hint Follow one ad request
Trace a single request from the moment a page or app asks for an ad to the moment one is shown. At each step ask what data is consulted and what decision it drives, then ask what happens to the data the impression produces.
```
### Clarifying Questions
- From which side should I answer: a publisher's ad server, an advertiser's buying platform, or an ad business that sells ads on its own marketplace and sites?
- Which formats matter most: display banners, sponsored product listings in search results, or affiliate placements?
- Is the focus the real-time choice for one impression, or also the offline work around it, such as audience building, forecasting and measurement?
### What a Strong Answer Covers
- The main kinds of data involved (about the user, the context, the advertiser's campaign, the creative and past performance) and where each comes from.
- Where each kind of data enters the serving path: eligibility and targeting, response prediction, auction and ranking, budget pacing, frequency limits and creative choice, all within a tight latency budget.
- The feedback loop from logged outcomes to attribution, model retraining and exploration.
- Constraints on data use: consent and privacy rules, the loss of cross-site identifiers, data freshness and cold start.
- How a product manager would tell whether data-driven serving is working for advertisers, the publisher and users.
### Follow-up Questions
- Third-party cookies and mobile advertising IDs become unavailable for most users. Which serving decisions degrade most, and what data replaces them?
- How would you test whether a new click-prediction model improves results for advertisers rather than just raising short-term clicks?
- A new advertiser has no performance history. How does the system decide when and where to show its ads?
Overview: A technical question for a product manager in advertising at Rakuten about how data shapes which ad is served for a given request. It tests understanding of targeting, response prediction, auctions, budget pacing, frequency limits, creative selection, privacy constraints and the measurement loop that feeds results back into serving.