First-Party Data on Streaming: What Amazon Knows That Netflix Does Not

Two different first-party graphs, two different jobs

Netflix knows what you watch. That's a genuinely rich signal for content-adjacent targeting and for understanding entertainment preference, and it's the foundation Netflix's own ad platform is built on. Amazon's authenticated graph is built from something different: account sign-ins, Fire TV device registrations and Prime Video streaming sessions, layered with actual purchase and browsing history from Amazon.com, reaching over 90% of US households deterministically. One graph knows what you're interested in watching; the other knows what you actually buy. Buying Netflix inventory through Amazon DSP, rather than through Netflix's own ad manager, is specifically what lets a campaign apply the second kind of signal to the first kind of inventory.

Neither graph is objectively better in the abstract — they answer different questions well. A viewing-behavior graph is strong at predicting what content someone will want to watch next; a commerce graph is strong at predicting what someone is close to buying. The value in this specific integration is having both available inside one buying decision, rather than being limited to whichever single graph the inventory's native platform happens to be built on.

The example Amazon itself uses, and why it generalizes

Amazon's own materials on its Disney integration describe a pet food brand targeting Disney content viewers who've separately shown interest in pet products on Amazon — a combination neither company's data could produce alone. The same logic applies to Netflix. A skincare brand buying Netflix inventory directly through Netflix's own platform can target broad demographic or content-affinity segments; the same brand buying the same Netflix inventory through Amazon DSP can target Netflix viewers who recently browsed a competing skincare product on Amazon, which is a commerce-intent signal Netflix's own viewing-behavior data structurally cannot produce on its own.

A worked example

Set a $10,000 Netflix line item through Amazon DSP against an audience built from Amazon browsing and purchase behavior — recent category shoppers, cart abandoners, or an in-market segment — rather than a broad demographic target. At a representative $30 CPM, that's roughly 333,000 impressions, and the honest expectation is the same top-of-funnel pattern as the rest of Amazon's streaming inventory: across 27 advertisers in our own DSP book over 31 days this summer, connected-TV and streaming overall returned 0.77x on strict last-click attribution with a 56.3% new-to-brand rate. What the Amazon-data layer changes isn't the attribution mechanics — it's who gets shown the ad in the first place, which is a targeting-quality question, not a measurement one.

The same principle scales down to a smaller test too. Even a $3,000 Netflix line item, too small to be a full campaign on its own, is worth pointing at a narrow, high-relevance Amazon audience rather than a broad demographic default — a small, well-targeted test teaches more than a larger, undifferentiated one at the same spend.

What to do when the targeting doesn't seem to be working

If a Netflix line item built around Amazon shopping signals isn't outperforming a broad demographic target, check audience size first — a narrow, high-intent Amazon-signal audience can be small enough that the campaign struggles to spend its budget efficiently against it, in which case widening the audience slightly, while keeping the commerce-relevance filter, is usually the right fix rather than abandoning the approach. If delivery is fine but performance still looks flat, confirm the audience segment is actually being applied to the Netflix line item and not defaulting to a broader, untargeted buy — a misconfigured audience setting is a common, quiet way this specific advantage never actually gets used.

Test one Amazon-signal audience against one broad demographic audience directly, at comparable budgets, before assuming the commerce-intent approach is automatically superior — the advantage is real in principle, but confirming it against a brand's own category and audience is worth the extra test rather than taking it on faith.

The common mistake

The mistake is buying Netflix through Amazon DSP and then running it with only broad demographic or content-context targeting, the same way it would run through Netflix's own platform — which throws away the entire reason to route the buy through Amazon DSP in the first place. If a brand isn't going to use Amazon's shopping and browsing signals to target the buy, there's a real argument that Netflix's own direct sales relationship might serve just as well, or better on placement guarantees. reMKTR builds every Netflix and Disney line item around an Amazon-signal audience specifically, because that's the differentiated value this routing exists to capture — running it undifferentiated is buying the same inventory the harder way.

The second version of the mistake is the opposite: assuming Amazon's data makes creative and content context irrelevant, and running the same commerce-focused audience without adapting the creative for a Netflix or Disney viewing environment. Better targeting doesn't substitute for creative that works in the actual viewing context — a household correctly identified as an in-market shopper still has to be reached by an ad that holds attention on a non-skippable, sound-on screen, or the targeting advantage is wasted on a creative that never lands.