How Amazon DSP Deduplicates Reach Across Streaming Services

The mechanism: authenticated, not modeled

Most cross-publisher reach measurement in advertising relies on probabilistic modeling — statistical inference about which impressions likely reached the same person, built from cookies, device fingerprints or panel data. Amazon's approach is different in kind, not just in accuracy. The authenticated graph is built from deterministic signals: an account sign-in, a Fire TV device registration, a Prime Video streaming session. These are direct, encrypted and pseudonymized connections to a verified household, not statistical inference, and Amazon's own published claim is over 90% of US households reached this way. Match rates on top of that base run above 75% on the open internet generally and above 80% specifically for streaming TV campaigns.

The graph also extends to Amazon's audio partnerships — Spotify, iHeart and Pandora — which matters for the streaming picture specifically because a household's identity, once authenticated, carries across screen and audio formats rather than resetting at each new device or app. For cross-publisher video specifically, the practical effect is that Amazon isn't inferring the connection between a Prime Video impression and a Hulu impression on the same TV — it's recognizing the same signed-in household or device on both, which is a materially stronger basis for both deduplication and frequency control than probabilistic matching provides.

What deduplication actually catches

Without deduplication, a household that saw an ad on Prime Video, Hulu and Netflix in the same week would count as three separate reach events in a naive sum of each platform's individual reporting — three "unique" viewers who are actually one person. The authenticated graph collapses that into a single household reached three times, which changes two numbers that matter for planning: your real unique reach (lower than the naive sum) and your real frequency (higher per household than any single platform's report would suggest on its own). Amazon's own reporting on this mechanism, applied to Interactive Video Ads specifically, cites 57% fewer redundant ad exposures as a result.

A worked example

Say a campaign runs $10,000 on Prime Video and $8,000 on Hulu in the same month, each reporting roughly 350,000 and 280,000 impressions respectively — 630,000 gross impressions if you simply added the two platform reports together. With deduplication applied through the authenticated graph, a meaningful share of those households appear in both platforms' delivery, and the real unique-household reach will land below 630,000, while frequency per household will be higher than either platform's individual average suggests. The exact overlap varies by audience and targeting, but the direction of the adjustment is always the same: naive addition overstates reach and understates frequency.

What to do when the numbers don't reconcile

If a campaign's deduplicated reach report looks implausibly close to the naive sum of its individual platform reports, check whether cross-publisher frequency capping is actually enabled for that campaign — deduplication and capping both depend on the same authenticated-graph configuration, and a campaign set up with independent per-platform caps won't benefit from either. If deduplicated reach looks lower than expected relative to spend, that can be a genuine sign of audience overlap concentration — the campaign may be reaching a smaller, more frequently-hit pool of households than the budget implies, worth checking against frequency distribution rather than assuming a reporting error.

If two different reports from the same campaign show different reach figures, check the reporting window and whether one report includes third-party supply sources like Netflix or Disney's DRAX bundle and the other only covers Amazon-owned inventory — a partial-scope report and a full-scope report will legitimately disagree, and that's a scope difference, not a data-quality problem.

The common mistake

The mistake is reporting a media plan's reach as the simple sum of each platform's individually-reported unique viewers — a number that's almost always too high once real household overlap is accounted for. It's an easy mistake to make because each individual platform's report is accurate on its own terms; the error is in adding them together without deduplication. reMKTR reports cross-publisher streaming reach off Amazon's deduplicated household figures specifically because a stakeholder who's shown an inflated reach number once will discount every number reported afterward, even the accurate ones.

The second version of the mistake happens in the other direction: treating a lower deduplicated reach figure as a sign the campaign underperformed, without recognizing that the naive, undeduplicated number it's being compared against was never real in the first place. A deduplicated number that looks smaller than expected is usually a more honest number, not a worse one — the fix, if the reach genuinely needs to grow, is broader targeting or added budget, not a return to inflated platform-by-platform addition.