Modeling Device Capabilities for Analytics

Netflix engineers Aarti Laddha, Richard Diaz-Cool, Rishika Idnani, and Venkatesh Selveraj describe how the company models device capabilities for analytics. The work sits against a platform that must support a large and changing mix of features and content types on many kinds of client devices.

The capability surface they call out spans 4K streaming, immersive audio, live streaming, and cloud gaming. Each of those features depends on different combinations of hardware, codecs, display, network, and software support, so a single “device” label is not enough for reliable analytics. Capability modeling turns that diversity into structured signals that can be joined to playback and product events.

For engineers building multi-device products, the practical problem is measurement: without a stable model of what a device can do, aggregates mix capable and incapable clients, feature adoption looks wrong, and A/B results get confounded by hardware and OS differences. Explicit capability dimensions make it possible to segment, filter, and attribute outcomes to real constraints rather than guessed device classes.

In streaming and interactive entertainment, competitors also ship 4K, richer audio, live events, and game-like experiences across TVs, mobiles, consoles, and browsers. Netflix’s public focus on device-capability analytics reflects a market where feature parity is less about launching a codec or a live pipeline and more about knowing, at scale, which clients can actually use each surface and how that changes over time.

Builders should treat device capability as a first-class analytics entity: define the dimensions that gate product features (resolution, audio path, live path, interactive or gaming path), keep them versioned as the device fleet evolves, and use them in every dashboard and experiment that claims feature success. Watch how new content types and client platforms expand that model so reporting does not lag the product.