Back to the directoryFind in the timeline
Netflix describes its recommender system and its business value
Two Netflix employees described the company's recommender, said it influences about 80 percent of streamed hours and estimated its value at over one billion dollars a year. These are company claims; the paper does not show how the estimate was made.
- Where it sits
- Consumer applicationUse of psychology, or claims about it, in commerce, policy or product design.
- Evidence level
- Company self-reportThe main evidence is the company’s own account. It has not been independently audited.
- Organizations
- NetflixAn organization is listed where the sources concern it. The text says exactly how.
- Sources inspected
- 2 of 21 with the relevant text, company document or archive item checked, 1 as an abstract or reference article, 0 not verified.
What happened
In December 2015 Carlos Gomez-Uribe and Neil Hunt, both Netflix employees, published a description of the Netflix recommender system in ACM Transactions on Management Information Systems. They describe a set of algorithms that build the homepage and search results. They say the system influences choice for about 80 percent of hours streamed, with the other 20 percent coming from search. They give the service as more than 65 million members.
The psychological idea
The paper states its business logic. Netflix says its revenue comes from monthly subscriptions that are easy to cancel, so keeping members is the main target. It reports that viewing time is strongly correlated with retention, so it tests algorithm changes in randomized A/B experiments lasting 2 to 6 months and compares streaming hours and cancellations. It also cites consumer research that members lose interest after perhaps 60 to 90 seconds of choosing.
In practice
The authors say a homepage has about 40 rows with up to 75 videos each, ordered by personalized rankers, and that personalization spreads viewing across more titles. They say years of development reduced monthly churn by several percentage points. They estimate that personalization and recommendations together save Netflix more than one billion dollars a year. The pages read do not show how this estimate was made.
Why it mattered
It is a first-party description of how a large streaming service says its homepage is built and what it says it optimises: member retention and medium-term engagement. It gives students a documented case of a platform stating that viewing time is the working measure it tests against.
Limits and qualifications
This is a self-report by Netflix employees, not an independent audit. The 80 percent figure says the system influences choice, not that viewers would not have found those titles otherwise. A later academic review notes that how Netflix's figures were estimated is not specified and that recommendations shown everywhere may be chosen simply because they are shown. The link between engagement and retention is Netflix's own correlation.
Evidence behind the claims
The authors state that Netflix had more than 65 million members streaming more than 100 million hours per day at the time of writing.
The authors say the recommender system influences choice for about 80 percent of hours streamed, with the remaining 20 percent coming from search.
The authors say they have reduced churn by several percentage points and think personalization and recommendations together save Netflix more than 1 billion dollars per year.
The authors report that time spent viewing is strongly correlated with retention and that A/B tests run for 2 to 6 months comparing engagement and cancellation across cells.
A typical homepage has about 40 rows with up to 75 videos per row, and the authors cite consumer research that a member loses interest after perhaps 60 to 90 seconds of choosing.
Jannach and Jugovac note that how Netflix's business-value figures were estimated is not specified in more detail, and that a mere presence effect can inflate adoption metrics for recommendations shown everywhere.
People involved
- Carlos A. Gomez-Uribe. Netflix, author of the 2015 paper
- Neil Hunt. Netflix chief product officer, author of the 2015 paper
In the Psy101 course
- OpenStax Psychology 2e, section 2.3: Analyzing FindingsChapter 2, Psychological Research. Method. The section discusses this topic.Opened the page: it has a heading stating that correlation does not indicate causation, which applies to Netflix's reported link between viewing time and retention.
This is course relevance only. It is not evidence for the historical or corporate claims on this page.See the full course map.
Sources and inspection
2 of 2 sources were opened and read at the level shown. A source marked as not verified supports no claim on this page by itself.
- [1]The Netflix Recommender System: Algorithms, Business Value, and InnovationGomez-Uribe, C; Hunt, N. ACM Transactions on Management Information Systems, 6(4), Article 13, 2015. Journal article. Company document.Company document readRead pages 1 to 10 of the 19-page PDF copy hosted by gwern.net (abstract, sections 1 to 4.2), which carries the ACM publication line December 2015. Written by Netflix employees. The ACM page itself was not opened.
- [2]Measuring the Business Value of Recommender SystemsJannach, D; Jugovac, M. arXiv 1908.08328 (ar5iv HTML version), 2019. Preprint. Critique or replication.Reference article readRead the first 100000 characters of the ar5iv HTML version as returned by the fetch tool, including Section 2.1 and its footnote 2 and Section 3.1.2; the final 7107 characters were not read. Authors confirmed from the arXiv abstract page.
Access note: The paper was read in a PDF copy hosted by gwern.net, not on the ACM site. The critique was read only in part (final 7107 characters unread).
Independent source review: Claude review agent, independent of the research agent, 2026-10-21. Verdict: pass-with-edits.