Netflix recommendation system
Improving personalization effectiveness.
Duration
7 weeks
Team
2 Product Designers
Role
Product Designer
Tools
Figma
Miro
Goal
Enhance Netflix's machine learning based content recommendation system to ensure tailored contents align closer to user preferences.
Problem
Preference-recommendation mismatch
Netflix's recommended content often not aligned with user preferences, leading to a passive viewing experience and distrust with the service quality.
Solution
I introduced features empowering users to actively engage with the algorithm. I designed touchpoints for users to remove disliked shows, swap out less-preferred genres, and customize recommendation setting that shift from a purely backend-driven approach to a user-in-the-loop experience.
x
Increase satisfaction score
%
Feature engagement rate
%
Increase in content discovery time
Scroll down for full story
Objective
Explore Netflix's machine learning recommendation system, identify areas for improvements and propose actionable solutions.
Research
54 Surveys
13 Interviews
Recommendation matching
How do users feel about the current recommendation experience?
New content discoverability
How enjoyable is it for users to discover new content?
Control and customization
Why do users feel specific control and customization option valuable?
Research discoveries
85%
Dissatisfied with Netflix recommendations
80%
Found Netflix discoverability limited
90%
Seek more control over algorithms' recommendations
"The system doesn't feel trustful. Everything pops up a perfect match, but in reality, I hardly liked them. Even I can't do anything about it."
Netflix user
Competitor analysis
We compared the relationship between user satisfaction with content discoverability and anatomy in competitors’ service to identify features Netflix can adapt to enhance personalization and rebuild trust.
Key insight
The passive recommendation experience limits user control, restricts content discoverability, and ultimately erodes trust through irrelevant suggestions.
How might we
How might we involve users in the personalization process to enhance the recommendation qualities?
Design goals
Integrative
To leverage user authority within Netflix's algorithmic strengths.
Flexible
To offer user a range of customization options that cater to diverse engagement preferences.
Convenient
To maintain ease of use with minimal disruption or added complexity.
Iterations and testings
Feature flowchart
The ideas sit on multiple product touch points so I looked at concept integration holistically in the flowchart, ensuring we were not only focusing on the feature but product experience as a whole.
Winning concepts
I focused on creating continuous touch points that allow users to input their preferences into Netflix's robust machine-learning algorithm, constantly keeping users in the loop.
Solutions
Show deprecation
Users can remove specific content from the recommendation list, their feedback will also been factored into the algorithm for future recommendation consideration. It solved the pain from micro level, allowing users to remove disliked content.
Collection swapping
Users can swap out entire genre or collections they dislike about, offering them more touch points to customize preferences with more feedback to the algorithm.
Algorithm calibration
Users can take holistic reboot or customization based of algorithm details. The transparency provide users with confidence and build up trust using the platform.
Outcome
Short and long term considerations
I introduced three user touchpoints to redesigned recommendation system: removing disliked shows, swapping unfavorable genres, and adjusting algorithm preferences. These features empower users to shape their recommendations while feeding valuable input back into the algorithm. This approach transforms passive recommendations into an interactive model, improving short-term personalization and fostering long-term trust and algorithmic learning.
Learnings and reflection
Uncovering the root cause leads to impactful solutions: Never stop to ask 'whys' to discoveries, how deep the problem understanding is drives the value of the solution.
Drive solution by maximum impact mindset: Impactful solutions don’t require revolutionary features; the key is maximizing value efficiently.