Device Capability Modeling at Netflix

Netflix serves a wide range of content — from 4K and immersive audio to live events and cloud gaming — across a heterogeneous device ecosystem. Hardware realities complicate this: RAM, CPU count, display support and platform constraints mean a given feature simply cannot run on every device model. To keep the experience consistent, the engineering team built a capability data model that merges device specifications with internal feature flag data. That foundation enables granular feature management, highlights penetration bottlenecks, and speeds up rollout decisions.

Storage and Data Structure

The core storage layer uses a cumulative table that tracks, for each device, its current state and associated capabilities — screen resolution, supported video profiles, surround sound, RAM size, and related attributes. This structure is optimized for analytics and reporting, as it efficiently captures the latest device state without needing to scan full history.

{
"Screen Height": ["720"],
"Screen Width": ["1280"],
"Video Profiles": 
[
"playready",
"hevc",
],
}

Aggregate Analysis with Histograms

For distribution-level questions, the team relies on a histogram table that counts active devices over the trailing 28 days, segmented by device model and software version. The table also records how many devices support specific capabilities, making it straightforward to compute penetration ratios. One practical application is tracking external display support on streaming sticks. In the example below, of the total devices counted, all support the HD profile (playready), while only 20 percent support the UHD profile (hevc).

{
"Video Profiles": {
      "playready": 100%, # HD profile
      "hevc": 20% # UHD profile
}
}

Analytical Products and Feature Reach

On top of these datasets, the team built analytical dashboards that give a full view of feature reach for initiatives like 4K Ultra HD, Netflix Spatial Audio, Cloud Gaming, and the latest UI. These data-driven views inform enablement decisions per device, helping to balance performance with feature adoption.