How to Evaluate a Household Graph for CTV: 5 Metrics That Matter

Connected TV keeps growing. IAB's 2026 Outlook forecast 13.8% growth in US CTV ad spend this year, and 72% of buyers named cross-platform measurement as a priority (StreamTV Insider). In IAB's later video report, buyers ranked targeting as their top spending criterion (49%), partly because IP-based signals are getting weaker (The Desk).

CTV comes with a unique identity challenge: the TV is a shared screen. To target, frequency-cap and measure CTV properly, you need to know which phones, laptops and tablets belong to the same household as the TV. That's what a household graph does. Here's how to judge whether one is good enough.

Match rate: how much of your audience can it see?

Start with the basics. Send a sample of your IDs (CTV device IDs, IPs, MAIDs or hashed emails) and ask what share the graph recognises. Check the result by country and by ID type, because a strong overall number can hide weak coverage in the markets you care about.

Household size: does the structure look like real life?

Count the devices per household. A healthy graph mostly shows households of 2–6 devices, with a long but thin tail. Warning signs are huge clusters of 30 or more devices, which usually means a shared IP such as an office, university or mobile carrier gateway was treated as a home. A big share of one-device households is another warning sign: it means the graph isn't connecting anything.

Precision: are the links correct?

Precision measures how many predicted links between devices are actually right. It's calculated as true positives ÷ (true positives + false positives), checked against a held-out set of deterministic data. That test data must never have been used to train the graph, or the results will look better than reality. For attribution, ask for a high-precision version of the graph. For reach, a lower threshold with higher recall can be the right trade.

Stability: do households hold together week to week?

IP addresses change and people move house. A graph that reshuffles households every week will break your frequency caps and muddy your attribution. Ask what share of households persists week over week. Our benchmark for most use cases is at least 90%.

Cross-screen connectivity: does it reach beyond the TV?

The real value of a household graph is connecting the living-room screen to the devices people carry. Ask what share of CTV households link to at least one mobile or desktop ID. That number decides whether you can retarget after a TV exposure, and whether you can attribute a website visit back to a CTV ad.

Putting it into practice

Run a test with two or three vendors on the same input file, and compare all five metrics side by side. A vendor that's top on match rate but weak on stability isn't a good deal. For more detail on testing methods, including overlap, deduplication and enrichment, read our white paper Evaluating Cross-Device Identity Graphs.

ROQAD's Household Graph links CTV, mobile and desktop IDs into households across North America and Europe. It's built on privacy-compliant signals, with consent tracked for every record.