The June report had a clean dashboard. Reach: 3.29M. Buyers: 34,032. Overall conversion rate: 1.0%. ROAS: 3.15.
Then someone isolated the DSP video line as a single touchpoint and got 0.2%.
That's the number that ends up in a QBR slide with a red arrow next to it. That's the number that gets a campaign killed with applause.
We almost let it.
The June report didn't lie. It just didn't tell the whole truth either.
We pulled actual customer journeys from Amazon Marketing Cloud. Not the summary Amazon hands you by default. The one you build yourself with SQL, covering 57 real paths over a 14-day window.
The pattern wasn't subtle:
Same audience. Same campaign. The only thing that changed was how many times someone saw an ad before buying.
Here's the thing nobody puts on the slide: last-click attribution isn't wrong. It's just answering a different question than the one you think you're asking. "What was the final thing someone saw before buying" is not the same as "did this ad do anything useful." Video ads are structurally bad at winning the first one on purpose. That's not their job.
We had built a report that graded an upper-funnel channel on a bottom-funnel test, then acted surprised when it failed.
Pausing the video line wouldn't have fixed the 0.2%. It would have removed the first touchpoint from every path that eventually converted at 18.3%.
When video was the first ad someone saw in their path, 66.2% of the buyers who came from that path were new to the brand. The account average new-to-brand rate was 49.8%.
Video wasn't failing to convert. It was finding people who'd never bought the brand before, warming them up, and handing them off to something else to close. Judging it on its own last-click CVR is like judging a doorman on how many people bought something at the register.
The doorman's job is to get people in the door. The register is someone else's problem.
This is the finding that actually changed the media plan.
Once we stopped looking at channels in isolation and looked at combinations:
Both of those are strong numbers. Neither exists in a report that only shows Sponsored Brands and Sponsored Display on their own, because the standard report has no concept of "assisted by."
The video line looked like the worst performer in the account. It was quietly a precondition for two of the best ones. Had we paused it, we'd have been staring at degrading Sponsored Brands performance the following month, running a completely different post-mortem with no idea why.
A note on what we didn't do: overstate what this analysis proves.
We excluded four rows below the aggregation threshold and said so in the report before presenting findings. Multi-touch paths will always look like they convert better than single-touch paths, partly because a shopper further along in a real buying decision is more likely to rack up multiple ad exposures in the first place. That's a selection effect. Touch three didn't necessarily cause the sale. People who were already going to buy tend to see more ads before they do it.
Being upfront about that cost us a slightly cleaner story. It was worth it. The moment you overstate what a path analysis proves, you hand the "this is all cherry-picked" objection to whoever wants an excuse to ignore it.
Here's what would have gotten this campaign paused, next to what saved it:
Video alone, last-click: 0.2% conversion rate. Video as first touch, full path: 66.2% new-to-brand rate. Video plus Sponsored Brands: 6.4% conversion rate, 3.7x ROAS. Video plus Sponsored Display: 14.1% conversion rate, 3.5x ROAS.
Same campaign. Same month of data. Two completely different verdicts, depending on whether you asked the report one question or five.
Amazon hands you the last-click report by default. Nobody sends you the path analysis. You have to go build it in AMC on purpose.
That means every account running on the default view is making an active choice, whether anyone frames it that way or not: measure the easy way and risk killing what's working, or measure the harder way and actually know.
Neither is dumb. But only one of them is honest about which one it is.
We set out to defend one video line item. What we came out with was a repeatable AMC path analysis and a real reason to stop reporting single-channel CVR without context. The video line is now one of the strongest new-customer engines in the account.
Still mediocre on last-click. Nobody's confused about why anymore.
The campaign didn't need saving. The report did.
The analysis behind this post is packaged as a Claude skill: an .md file containing the AMC SQL query, the logic for structuring path data by touch count and channel combination, and notes on exclusion thresholds and caveats to include when you report it out. You run it against your own AMC instance, against your own campaign data, and get back the same output we walked through here. It's not a one-time exercise we ran for one account. It's a repeatable query any team with AMC access can use before the next QBR puts a number on trial.
Want the AMC SQL behind this analysis? Reach out at hello@bidx.io or book a call and we'll send you the .md file.