Skip to content
Case StudyExternal AdsApril 2026

A 74% ACoS test looked like a failure. Non-attributed weekly sales doubled

Claude read past surface metrics to find the late conversions and organic-rank lift hiding inside an off-Amazon influencer test.

Start free trialSee external ads tracking

2 weeks free · no credit card

propamp.ai / external-ads / verdict-lab
0%
External ACoS that made the test look bad
0.0%
tACoS before the read
0.0%
tACoS after organic lift surfaced
0 min
wrap-up once the agent had the data

As featured in

  • Forbes
  • Tech Times
  • The AI Journal
  • Brainz

The numbers that flipped the verdict

The losing ACoS metric was not the whole profit story.

The losing ACoS metric was not the whole profit story.

The campaign looked expensive in isolation. Once Claude connected total ACoS movement and non-attributed sales, the test changed from kill to keep learning.

tACoS

0.0%
tACoS before
0.0%
tACoS after
lower total ad load

Non-attributed sales

$0/wk
Non-attributed sales before
$0/wk
Non-attributed sales after
weekly sales doubled

Surface read

74% External ACoSorganic lift foundverdict flipped

Contrarian metric

The 74% External ACoS stayed visible, but it stopped being the only number allowed to judge the test.

The test

Four influencer placements for a $150 smart dog collar.

Four influencer placements for a $150 smart dog collar.

A solo six-figure seller ran the four-week experiment in a competitive niche, tagged through Amazon Attribution across the ad sources that mattered.

0
weeks

April 2026 test window

0
placements

Influencer pushes tagged through Amazon Attribution

$0
product

Smart dog collar in a competitive niche

6-fig

seller

Solo private-label operator running the test

campaigns / attribution / source-map
Amazon Attributiontagged
Googlesource spend
Metasource spend
Influencers4 placements

Why surface metrics lied

Two ROI drivers were hiding outside the campaign report.

Two ROI drivers were hiding outside the campaign report.

The agent treated the campaign as one input in the product system, not a standalone ad console.

14-day attribution window

Late conversions escaped the 14-day window

The campaign made demand that did not land inside the attribution report, so the ad test looked more expensive than it was.

organic-rank lift

Velocity pushed organic rank up

The traffic spike helped the listing climb. The sales it created were then labeled organic instead of ad-driven.

The read in the seller's words

The real win was not visible in the first report.

The real win was not visible in the first report.

agent / attribution-thread
1Read attribution data
2Check non-attributed sales
3Compare organic rank
"A"A'failing''failing'off-Amazonoff-Amazonadadtesttestboostedboostedthethelisting'slisting'sorganicorganicranking."ranking."

The wrap-up took about 10 minutes once the agent had the data.

External ads case note

What comes next

The seller did not stop the channel. They tightened the measurement.

The seller did not stop the channel. They tightened the measurement.

Three more tests are now being run with keyword rank tracking wired in, so the organic lift is measured directly instead of inferred after the fact.

The next experiment starts with the missing signal attached.

External ACoS still matters. It just no longer gets to ignore late conversion behavior and organic rank movement.

01

3 more off-Amazon tests queued

02

Keyword rank tracking wired in

03

Organic lift measured directly

See external ads trackingSee Metrics

Your move

Start your free PROPAMP AI trial.

Start your free PROPAMP AI trial.

Fold every ad dollar into product profit, then let an agent check whether the surface metric is telling the whole story.

Start free trialBack to case studies
sku-profit / external-spend / rank-lift
SpendSalesRank