Upper-Funnel Measurement Without a Click
Upper-funnel formats — streaming, connected TV, display, in-app video — are built to be seen, not clicked, so judging them on click-based metrics misreads them by design. The honest alternative is a stack of proxy measures: video completion rate, viewable impressions, Brand Lift surveys, and geo-tested branded search lift — used together, since no single one proves incrementality alone.
What this looks like in a real account
Why click-based metrics were never going to work here
A streaming ad plays on a television screen with no clickable surface at all in most environments. A connected-TV or in-stream video ad interrupts a passive viewing experience where the expected action is watching, not tapping. Judging either on click-through rate isn't measuring performance badly — it's measuring the wrong thing entirely, because the format was never built to produce the interaction the metric assumes. This isn't a new problem specific to Amazon; it's the reason television advertising has always used different measurement than direct-response formats, and the same logic applies once that inventory is bought programmatically through Amazon DSP.
What actually is measurable, without a click
Four proxy measures do real work here. Video completion rate tells you whether people who started watching stayed through the message — a low completion rate on a 15 or 30-second spot is a creative problem worth fixing before anything else. Viewable impressions, measured against Media Rating Council standards, confirm the ad was genuinely in view rather than served but never seen. Brand Lift studies — available for DSP and most Sponsored Ads formats — survey exposed versus control audiences on awareness, favorability and purchase intent, giving a genuine read on whether perception moved. And geo-tested branded search lift, run as a proper matched-market test rather than a simple before-and-after comparison, checks whether the upper-funnel exposure is producing the downstream search behaviour it's usually meant to spark.
A worked example of what this looks like together
In our own DSP book, streaming and connected-TV inventory returned a 0.77x ROAS — the worst return multiple on the whole report — while carrying the highest new-to-brand rate of any inventory type, 56.3%, on $33,525 of streaming spend at a $15.51 CPM against $3.71 portfolio-wide. Read on ROAS alone, that line looks like the first thing to cut. Read through the proxy stack instead: if completion rate is strong, viewability is clean, a Brand Lift study shows measurable awareness gain, and a geo-tested branded search comparison shows a real lift in test markets versus control — that's four independent signals, none of them a click-based sale, all pointing the same direction. That combination is a genuinely stronger case for the spend than a same-session ROAS number was ever going to provide, because the format was never built to earn one.
Why our own DSP practice sets streaming its own target rather than a shared one
Our standing practice is to set streaming a return target near 1.0 and judge it primarily on completion rate and branded-search lift, because attribution structurally undercounts it — Amazon's shoppable features on streaming inventory generally only fire for logged-in Prime members, so a meaningful share of streaming-driven demand never generates an attributable click or sale on Amazon at all. Holding streaming to the same ROAS bar as retargeting, which converts demand that already exists, guarantees the streaming line loses every comparison regardless of how well it's actually working — which is exactly the trap a shared blended target sets.
The common mistake, including ours
The mistake is defaulting to the same efficiency target across every placement type, then being surprised when upper-funnel formats consistently underperform against it. We've inherited accounts where a client's prior reporting held streaming to the account's blended ROAS target, month after month, and the resulting recommendation — every single time — was to cut streaming spend, because no upper-funnel format was ever going to clear a bar calibrated to same-session retargeting performance. The spend wasn't necessarily failing; the target was measuring the wrong thing against it from the start.
Building the case incrementally, not all at once
None of these four proxy measures needs to be running simultaneously from day one. A reasonable build order is completion rate and viewability first, since they're available in standard reporting from the first flight and tell you quickly whether the creative itself is landing; a Brand Lift study next, once the campaign has run long enough to reach eligibility thresholds; and a geo-tested branded search lift last, once you have a stable baseline period to match test markets against. Trying to stand up all four before a single flight has run usually means none of them are done well.
When the proxy signals disagree with each other
If completion rate and viewability look strong but Brand Lift shows minimal movement, treat that as a real signal the creative or targeting needs work, not as a reason to distrust the survey. If Brand Lift shows a positive result but a geo-tested branded search lift comes back flat, check test power before concluding the two genuinely disagree — branded search volume is often a smaller absolute number than perception-survey responses, and an underpowered geo test can show no lift even when a real one exists. Only after checking power should a genuine disagreement between the signals change your read on the channel.
| Proxy metric | What it tells you | What it doesn't |
|---|---|---|
| Video completion rate | Whether the creative held attention through the message | Whether it changed behaviour afterward |
| Viewable impressions | Whether the ad was genuinely seen (MRC standard) | Whether being seen mattered |
| Brand Lift survey | Whether perception moved for the exposed group | Whether the perception shift produced a sale |
| Geo-tested branded search lift | A causal signal on downstream search behaviour | Everything happening beyond that specific outcome metric |
Which one you should actually pick
Any advertiser running DSP or Sponsored Ads video and display can pull completion rate and viewability from native reporting, and request a Brand Lift study directly from Amazon at no extra platform cost. Where reMKTR's practice adds value is setting a genuinely separate target for upper-funnel spend before the first impression serves, rather than measuring it against a bar it was never built to clear — part of the discipline behind Full Circle's $500M+ in managed Amazon spend across 100+ brands.
Shortlist on the job, not the feature grid. Pull your search-term report for the last 90 days and total the spend against terms that produced no orders — 33.6% on the account above. Then ask each vendor on your list what they would do about it in week one, and see who answers with a process rather than a screenshot.
Common questions
Should upper-funnel media have the same ROAS target as retargeting?
No — retargeting converts demand that already exists and will almost always show a stronger same-session return. Set upper-funnel formats their own target, informed by completion rate and search lift rather than ROAS alone.
Is a low ROAS on streaming or connected TV always a bad sign?
Not by itself — check completion rate, viewability, Brand Lift and branded search lift before concluding the spend isn't working. A format built to be seen, not clicked, will structurally underperform on click-based measurement regardless of how well it's actually doing.
Why does streaming attribution undercount real performance?
Shoppable features on streaming inventory generally only fire for logged-in Prime members, so a meaningful share of streaming-driven demand never generates an attributable action on Amazon at all, even when the ad genuinely influenced a later purchase.
How many of these proxy metrics do I need before trusting a result?
More than one, ideally — a single metric in isolation is easier to misread. Two or more proxy signals pointing the same direction is a meaningfully stronger case than any one number alone.
We show the method before the number.
Claim the free auditRead next
- Acorn Cost: What Decides the Invoice, and What to AskPricing · acorn cost
- Acorn vs Podean: Two Amazon Agencies ComparedHead to head · acorn vs podean
- Pacvue Pricing: No Public Number — What to AskPricing · pacvue pricing
- Skai vs Pacvue: Contracts, Not Feature GridsHead to head · skai vs pacvue