United StatesMedia and entertainment2009 – ongoingNetflix
Recommendation and personalization system
Recommendations drove about 80% of Netflix viewing hours, and Netflix valued them at over $1 billion a year
Problem
A streaming subscriber who cannot quickly find something to watch may stop using the service and cancel. Netflix's catalog was too large for members to browse, and each lost subscriber had to be replaced with a new one.
Approach
Netflix built a set of in-house algorithms that rank titles on the home page, form personalized rows, pick similar titles, choose artwork and power search. The company moved from predicting star ratings, the focus of its DVD era and the Netflix Prize, to models based on streaming behavior. It tested changes through A/B tests that measured retention and medium-term engagement.
Results
Takeaway
Netflix valued recommendations by the cancellations they prevented, without measuring direct sales. The $1 billion figure is Netflix's own 2015 estimate.
Categories
- Size
- Large enterprise
- Industry
- Media and entertainment
- Business function
- Marketing and salesIT and R&D
- Innovation type
- Product and service innovation
- AI technology
- Recommendation and personalization
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
Carlos Gomez-Uribe and Neil Hunt of Netflix describe the algorithms behind the service's recommendations and how the company tests them with A/B experiments. They write that recommendations influence about 80% of hours streamed, with the remaining 20% coming from search. They state that personalization has reduced monthly churn by several percentage points and estimate the combined effect at more than $1 billion a year.
Tubefilter reports the $1 billion yearly estimate by two Netflix executives and explains that the savings come mainly from retaining subscribers. The writer notes that the estimate cannot be backed up with hard numbers and may be somewhat overstated.
Full details are in the original sources.
