What is a device graph and how does it work in CTV advertising?

A device graph is a database that links multiple devices to the same household or individual. In digital advertising, a person typically owns several devices — a smartphone, a laptop, a tablet, and a smart TV. A device graph connects these devices so that an ad system can recognise that the Android TV in the living room and the iPhone in the viewer's pocket belong to the same household. This connection enables cross-device frequency capping (limiting total impressions across all household devices), cross-device attribution (connecting a CTV ad impression to a mobile app install), and audience targeting that applies data from one device to ads on another. Device graphs are the infrastructure layer that makes multi-device advertising coherent.

India CTV device graphs are built primarily through deterministic and probabilistic matching. Deterministic matching uses a shared login identifier — the same Google account signed into an Android TV and an Android phone links those devices with high confidence. Probabilistic matching uses household IP addresses: devices sharing the same home broadband IP at the same times are inferred to be in the same household. DSPs like DV360 and The Trade Desk maintain their own device graphs for India, primarily built on Google identity (deterministic) and IP-based household matching (probabilistic). Telecom-based device graphs — linking Jio subscriber data across Jio broadband and JioHotstar — are more accurate but are not exposed to the open programmatic market.

Three India-specific factors reduce device graph accuracy: (1) Dynamic IP addresses — India's major ISPs (Jio Fiber, Airtel, BSNL) frequently reassign broadband IP addresses. A household's IP address may change every few days, breaking the IP-based household link used for probabilistic device matching. (2) Mobile-first internet usage — many Indian households primarily access the internet through mobile data rather than home broadband, reducing the shared home IP signal that probabilistic graphs depend on. (3) Multi-dwelling buildings — apartments in Indian cities often share IP ranges, making IP-based household separation less reliable in dense urban environments. These factors result in effective cross-device match rates of 20–40% for India CTV-to-mobile connections, compared to 40–65% in the US or UK.