Netflix Churn And Ip Sniffing

Here is a number that should get your attention: $1 billion.

That is roughly how much Netflix was losing annually -- not to competition, not to bad content, but because users were opening the app, scrolling for a few minutes, and closing it without watching anything.

If a user pays for Netflix every month but never watches, they have one predictable reaction at renewal: why am I paying for something I never use? At scale across hundreds of millions of subscribers, that question translates to a churn figure worth a billion dollars a year.

Why 5 Million Choices Is a Problem

As of 2016, Netflix had approximately 5 million titles -- movies, documentaries, series across dozens of languages and every genre imaginable.

That scale is the problem. Open Netflix and see five titles in languages you do not understand, and you close the app. Open it and see content that bears no resemblance to anything you have watched, and you close the app.

The right frame here is not "we need better content." Netflix had content people wanted to watch. The problem was that users were not finding it.

The product was failing at one specific job: matching the right content to the right user in the time it took them to decide whether to stay or leave.

This is a product management problem. The PM's first contribution was not a solution -- it was identifying the right metric to look at: sessions without content plays.

Out of thousands of data points a team could examine, this one turned out to be the leading indicator of churn. Users who opened the app and browsed without clicking on anything were the users most likely to cancel their subscription. That single observation, reached through data analysis and user interviews, is what made the rest of the work possible.

The IP Sniffing Solution

The insight behind the solution came from an offline observation: people in the same city, speaking the same language, moving in similar social circles tend to enjoy similar content.

If a colleague watched something over the weekend and could not stop talking about it, the probability is high that you would also enjoy it.

Netflix operationalized this without ever asking users for location data. Asking for location permission introduces friction -- some users decline, some distrust it, some simply find it annoying. Any barrier that causes users to abandon onboarding is itself a product failure.

Instead, Netflix used IP sniffing. IP addresses carry approximate location information. Without requesting location permission, Netflix could infer roughly where a user was accessing from and use that inference to surface content trending among users in the same geographic cluster, filtered by language and watch history.

ExpandProduct manager frames the problem from data; developer finds the frictionless solution -- together they drive measurable outcomes

The result was the "Top 10 trending in India" and "Top 10 trending near you" widgets that appeared on Netflix in 2016. After launch, the percentage of users unsubscribing each month fell by 4%.

Across 100 million subscribers, that is 4 million people who stayed. The annual savings translated to $1 billion.

Why IP Sniffing Beat Every Other Option

The product manager would have considered several alternatives: ask users to select language preferences during onboarding, request location permission explicitly, use a questionnaire to gather genre preferences.

Each of these carries friction. The rule worth internalizing: whenever you build for users, assume they are lazy.

If you make users do too many things, most will not do it. UPI became the default payment method in India because scanning a QR code is faster and less effortful than anything that came before it. Reels does not ask what kind of content you want -- it infers it. The products that win are almost always the ones that demand the least from the user.

IP sniffing required zero user action. It updated automatically as users traveled. It was accurate enough to be useful without being precise enough to feel invasive.

The developer's contribution was finding the implementation that was recurring, accurate, and completely frictionless -- those three constraints together ruled out every simpler approach.

There is a regulatory note worth keeping in mind. In Europe, collecting data about users without explicit consent is tightly restricted, and IP-based location inference sits in a gray area legally. As developers in product companies, awareness of data regulation is part of building responsibly.

What Each Role Contributed

The product manager spent months digging into why churn was high. They identified the behavioral pattern, formed the hypothesis, and brought the problem statement to the developer: users who do not watch content do not renew.

The developer's job was not to pick from a list. It was to reason about the trade-offs of each approach and land on the one that was technically accurate, scalable, and required nothing of the user.

Neither role could produce this outcome alone.

The PM without the developer has an idea that never ships. The developer without the PM has no way of knowing which metric to chase or why it matters. When both roles do their work well, the result is a feature that saves $1 billion.

The next post in this series looks at three more products that faced the same core problem -- users arriving but not converting -- and how behavioral nudges, A/B testing, and data-driven iteration solved each one.

The Essentials

1. Sessions without content plays was the metric that unlocked Netflix's solution -- identifying the right data point out of thousands is itself a PM skill.
2. IP sniffing beats location permission because it requires zero user action. The frictionless solution almost always wins over the comprehensive one.
3. The PM frames the problem. The developer finds the solution. Neither produces the outcome alone. The billion-dollar result required both working in sequence.

Further Reading and Watching

• Netflix Recommendations: Beyond the 5 Stars (Netflix Tech Blog) -- the technical and behavioral foundations of the recommendation engine discussed here
• Inside Netflix's Recommendation Algorithm (Wired) -- how Netflix used viewing data and geography signals to solve the content discovery problem
• Good Strategy / Bad Strategy (a16z, YouTube) -- on why framing the right problem is more valuable than jumping to solutions