Twenty dollars and twelve hundred dollars are reported CAC numbers for the same YouTube campaign, measured the same week, off the same spend. The twenty-dollar figure came from engaged-view credit. The twelve-hundred-dollar figure came from counting only the conversions that followed an actual click. Same budget, same creative, a sixty-fold gap in what the campaign appeared to cost per new customer, depending on which column of the report someone chose to read.
That gap is not a tracking bug. It is Demand Gen attribution working exactly as designed, and most ecommerce teams scaling budget into YouTube, Shorts, Discover, and Gmail campaigns have never seen the two numbers side by side.
- Demand Gen campaigns run on YouTube, YouTube Shorts, Discover, and Gmail, and Google leans on non-click signals to measure them because most viewers never click a video or image ad before buying.
- An engaged-view conversion counts someone who watched a chunk of the ad, clicked nothing, and purchased within a day: Demand Gen's default lookback window is shorter than the 3-day default used for standard web conversions.
- That counting method can make a campaign's reported cost per acquisition look dramatically better than its click-only cost, without the campaign necessarily deserving the credit.
- The fix is not disabling the metric. It is reading click-through CAC, view-based CAC, and an incrementality check side by side before a budget decision gets made on any one of them alone.
What a Demand Gen campaign actually is
Demand Gen replaced and absorbed what used to be called Discovery ads and Video Action campaigns. Google's own documentation on the Video Action campaign upgrade confirms that existing Video Action campaigns were migrated into Demand Gen rather than run as a separate format going forward, which means any ecommerce brand still reading an old Video Action campaign guide is reading about a format Google has already folded into this one.
The pitch is straightforward: one campaign type, AI-selected creative combinations, served across YouTube in-stream, YouTube Shorts, Discover feeds, and Gmail promotions tabs. Google's documentation on Demand Gen metrics and reporting describes it as built for visually-led, upper-funnel demand creation rather than for capturing an existing search intent the way Search or Shopping campaigns do.
That upper-funnel framing is also the root of the attribution problem. A Search ad gets clicked, then the sale happens, and the connection is direct. A Demand Gen video gets watched mid-scroll, nobody clicks anything, and the viewer buys later that same day after searching the brand name directly or just typing the URL in. Click-based attribution has nothing to attach that sale to. Google built a different counting method specifically to fill that gap.
What engaged-view conversions actually count
An engaged view, per Google's own support page on engaged-view conversions, is counted when a viewer watches at least 10 seconds of a skippable in-stream ad, or the whole ad if it runs shorter than 10 seconds, watches at least 5 seconds of an in-feed or Shorts ad, clicks nothing at any point, and then converts on the site inside the attribution window. The same documentation lists a 3-day default lookback for standard web conversions, but Video Action, Demand Gen, and Performance Max conversion actions default to a tighter 1-day window instead.
No click is required anywhere in that sequence. View-through conversions are the broader version of the same idea: any view, not just one that clears the engagement threshold, followed by a conversion with no click in between. Demand Gen attribution draws on both, plus Google's data-driven attribution model, and credits conversions across YouTube, Shorts, Discover, and Gmail using signals that exist only inside Google's own ecosystem.
That is the mechanical reason the twenty-dollar CAC number exists. The campaign gets credit for anyone who scrolled past the ad, watched enough of it to clear the threshold, and bought within that same day, regardless of what actually convinced them to buy. A shopper who had already decided to purchase after reading reviews elsewhere, who happens to also watch 10 seconds of a retargeting video a few hours before checking out, hands the campaign full engaged-view credit for a sale it did not meaningfully influence.
Reading the real number
The practical fix is a three-way comparison, not a setting to toggle off. Pull the campaign's conversions broken into click-through, view-through, and engaged-view, then calculate CAC for each segment separately instead of trusting the blended total Google Ads shows by default. A campaign where click-through CAC and the blended CAC are reasonably close is earning its credit honestly. A campaign where the gap runs into multiples, the way the twenty-versus-twelve-hundred-dollar example did, is mostly getting credit for demand it did not create.
The same logic that applies to a store's blended MER and new-customer CAC applies here: a single platform-reported number was never going to be the full picture, and the discipline is reconciling it against a second, more conservative read before a budget decision leans on it.
Once the click-only CAC is in hand, the next question is whether the campaign is actually incremental, meaning whether those sales would not have happened without it. View-based attribution cannot answer that on its own, no matter how the window is set. A short, deliberate incrementality test, holding back spend to a comparable audience for a defined period and comparing the lift, is the only way to separate a campaign that creates demand from one that is simply standing in front of demand that already existed and taking the credit.
When to scale, and when to hold
If click-through CAC sits within a reasonable range of the blended figure, and a holdout test shows a real lift versus the control, that is a campaign earning the right to more budget. Scale it the way any working channel gets scaled, in deliberate increments, watching cost per acquisition as spend increases rather than assuming the historical blended number holds at a larger scale.
Budget pacing matters here more than it does on Search. A Demand Gen or video campaign generally needs a daily budget several multiples of its target cost per acquisition before Google's delivery system has enough signal to optimize reliably, so a campaign funded at only one or two times its target CPA per day will look noisy and underpowered regardless of how good the creative is. Underfunding a test is one of the quieter reasons a channel gets written off as "not working" when the real problem was never giving it enough daily signal to learn from.
If the gap between click-through and blended CAC is large and no incrementality test has ever been run, the honest move is to treat the campaign as an unproven upper-funnel bet, not a performance channel, until one of those two things changes. That does not necessarily mean cutting the budget. It means budgeting it like exploratory spend rather than like a lever with a confirmed return, and revisiting the decision once an actual holdout test has run.
A reporting setup that cannot break out click-through, view-through, and engaged-view CAC separately, or that has no record of an incrementality test ever being run on a scaled channel, is a measurement gap worth closing before the next budget increase, not after it. That is the exact kind of discrepancy a measurement-led ads management engagement is built to catch before spend scales into it, and it is also a question a marketing analytics audit can answer directly if the accounts are currently run elsewhere and the numbers just need a second opinion before anyone trusts them.
Pull the two CAC numbers side by side before the next budget conversation. The gap between them, not either number alone, is what actually tells you whether the campaign is working.
