Household Targeting on Connected TV

Why household-level, and why deterministic

A connected TV is typically a shared device — the whole point of it is that it's the household's screen, not one person's phone. Targeting it meaningfully requires recognizing the household, not chasing an individual cookie the way display targeting on the open web historically has. Amazon's authenticated graph does this deterministically: built from account sign-ins, Fire TV device registrations and Prime Video streaming sessions, encrypted and pseudonymized, reaching over 90% of US households. Match rates run above 75% on the open internet generally and above 80% specifically for streaming TV campaigns — a materially stronger foundation than the probabilistic, cookie-and-fingerprint-based matching that dominates much of the rest of programmatic CTV.

The deterministic foundation also matters for a brand operating in a privacy-conscious category. Because the graph runs on encrypted, pseudonymized signals rather than raw personal data changing hands, targeting can stay precise without the brand itself ever seeing or handling individual customer identities — the household match happens inside Amazon's own systems, not in a data file handed to the advertiser.

What this actually enables

The practical output is Amazon's more than 20,000 internal audiences, built on real purchase, browsing and streaming signals rather than inferred interest categories — and the ability to apply those audiences at the household level on a CTV screen. In our own accounts, the single highest-converting audience we work with is households that added a product to cart and didn't buy, including ones who'd never purchased from the brand before. A large share of cart abandonment on Amazon is interruption, not rejection — the session ended, the app closed, the moment passed — and retargeting that household on a connected-TV screen days later recovers demand that already existed rather than trying to manufacture new intent from a cold audience.

This same household-level foundation is what makes cross-device retargeting genuinely coherent rather than a loose approximation. A shopper who browsed a product on their phone at lunch and abandoned a cart can be retargeted that evening on the living-room TV, because both events tie back to the same authenticated household rather than to two disconnected devices a probabilistic system would have to guess were related.

A worked example

A campaign built around cart-abandoner retargeting on Prime Video and Fire TV, at $8,000 for the month against a smaller, warmer audience than a broad prospecting buy, will generally deliver fewer total impressions but a meaningfully lower cost per outcome. Across 27 advertisers in our own DSP book over a 31-day window this summer, retargeting cost $4.47 per attributed customer against $12.13 for prospecting on roughly equal spend — prospecting produced a higher new-to-brand rate, 46.6% against 29.2%, which is the actual reason to run both rather than picking one, but on cost per outcome alone, retargeting wins by a wide margin. Pair that retargeting line with a separate, larger prospecting budget targeted by category-relevant in-market audiences, and the household graph lets both run under one shared frequency cap rather than risking the same household getting hit by both simultaneously.

What to do when household targeting isn't converting

If a cart-abandoner retargeting line item isn't converting at the rate you'd expect, check the recency window first — retargeting someone who abandoned a cart six weeks ago is a much colder audience than someone who abandoned it six days ago, and a window set too wide dilutes what should be the warmest segment in the account. If a health or supplement brand's retargeting audience is thin, remember the constraint on this category sits on unproven claims in the creative, not on the audience itself — a common misunderstanding that leaves the highest-converting segment completely unworked for brands that assume the category is blocked outright.

If retargeting reach looks smaller than expected relative to your overall Amazon traffic, check the underlying pool size before assuming the targeting is broken — a brand with genuinely low cart-abandonment volume on Amazon simply has a smaller retargeting audience available, and no configuration change increases a pool that isn't there. In that case, the honest fix is building the top of the funnel first, not tightening or loosening retargeting settings against an audience that doesn't yet exist at scale.

The common mistake

The mistake is building a CTV plan entirely around broad prospecting audiences and skipping household retargeting because it feels like a smaller, less impressive line item on a media plan. It's smaller in reach, and it's frequently the most cost-efficient line in the whole account — the two facts aren't in tension, they're the actual argument for running it. reMKTR runs 109 live Amazon DSP advertiser seats and builds a retargeting line into every CTV plan from the start, sized to the account's actual pool of recent cart abandoners rather than treated as an afterthought layered on once prospecting is already funded.

The opposite mistake also shows up: running retargeting alone and skipping prospecting because retargeting's cost per outcome looks so much better on paper. A retargeting-only plan eventually exhausts its own audience pool — it can only ever reach people who already interacted with the brand, and without prospecting feeding new households into that pool, the whole plan quietly shrinks over time even as its unit economics keep looking strong.