Explain how Amazon's recommendation engine picks the products each shopper sees
Company: Amazon
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
In a product manager interview at Amazon, the interviewer asks a technical question: explain how product recommendations work on Amazon. Aim for the depth a PM needs to work credibly with the engineers and scientists who build the system: where shoppers encounter recommendations, what data feeds them, how the system decides which products to show and in what order, and how the team can tell whether the recommendations are doing their job.
```hint Start from what the shopper sees
Pick two or three places where a shopper meets recommendations, such as a product page and the home page, and ask what question each one answers for that shopper. The data and the logic follow from that question.
```
### Clarifying Questions
- Should the explanation cover one surface, such as the product page, the home page, the cart or recommendation emails, or recommendations across the whole shopping journey?
- How technical should it be: the concepts a PM uses with the science team, or the specific model families and their trade-offs?
- Are sponsored product placements, which are paid ads, in scope, or only organic recommendations?
- Does the interviewer also want to hear how recommendation quality is measured and improved, or only how the system works today?
### What a Strong Answer Covers
- The recommendation surfaces a shopper meets and the different shopper intent each one serves
- The shopper, product and context data the recommendations learn from, and the biases built into that data
- How the system narrows a huge catalog to a short, ordered list fast enough for a page load, including what happens for new shoppers and new products
- How recommendation quality is judged, both before launch and on live traffic, and which guardrails apply
- The product decisions a PM owns, such as what the system optimizes for and how it handles privacy and shopper control
### Follow-up Questions
- A newly listed product has no views or purchases yet. How does it get recommended at all, and how would you check that the approach is fair to new sellers?
- After a model change, clicks on the bought-together bundle rose but revenue per visitor fell. How would you investigate?
- If shoppers said the recommendations felt invasive, which signals or surfaces would you change first, and what would that cost?
Overview: A product manager interview question that asks the candidate to explain how Amazon's product recommendations work. It tests knowledge of recommendation surfaces, shopper and item signals, collaborative filtering, candidate generation and ranking, cold start for new shoppers and products, and measuring recommendation quality with experiments.
