Claude spotted that a "failing" off-Amazon ad test boosted the listing's organic ranking
helping an FBA private-label seller scale in a competitive niche
May 2026 · 7 min read
Reading off-Amazon ad results with Claude Code
Mid-May 2026. A US seller — solo operator, 6-figure annual revenue — sat down to review his first off-Amazon advertising test. Four weeks through April: four placements, four different dog creators on Instagram, TikTok, and YouTube. The product: a smart GPS + health-tracking dog collar at $150.
He runs his Amazon ads himself — no agency, no PPC manager on staff. His Amazon campaigns are dialed in: top keywords captured, budgets allocated where they convert, target ACoS holding. But every extra dollar on Amazon now brings less back. He wanted to see whether a different channel could move units that his Amazon ads no longer could.
Data on the platform
PROPAMP AI is a next-gen inventory management and profit tracking platform.
All four influencer campaigns were on the platform from day one — logged as Manual Campaigns, tagged through Amazon Attribution, with clicks and sales feeding in automatically. Internal Amazon ads, Google ads, Meta ads, and influencer placements all line up in the same reports. (TikTok ads integration is coming soon)
Wrapping up the test
The seller works with Claude Code — Anthropic's AI agent that runs locally on his computer with access to read and edit his files. Every test he runs gets documented in the chat — assumptions going in, the data, what he decided afterward. The agent carries the context forward.
When the seller sat down to close out the influencer test, he opened the conversation:
Seller:
"pull the influencer test data from propamp. looks like a bust - ads aren't paying off"
The agent connected to the PROPAMP AI MCP server (Model Context Protocol — what lets an AI agent pull data directly from connected software), pulled the campaign data, and came back:
Claude Code:
"Pulled the data. The 74% External ACoS is accurate — that's direct attribution inside Amazon's 14-day window. But tACoS on this product dropped over the test month, and nothing else changed — same Amazon campaigns, same prices. My read: the influencer traffic is producing two effects that don't show up in External ACoS. First, late conversions outside the 14-day window. Second, the higher sales velocity during the test feeds Amazon's ranking, lifting organic sales. Both land in non-attributed sales, and both show up in total tACoS. Worth verifying week by week before closing this out."
Weekly breakdown Claude shared with the seller in chat. Click to expand full image
The seller opened the Statistics and Metrics report inside PROPAMP AI to verify. One click on weekly granularity. There it was — Internal, External, and Total ACoS and tACoS, side by side. The check took under a minute.
tACoS dropped from 14.0% to 10.7%. Non-attributed sales — Total revenue minus Amazon Ads sales minus off-Amazon attributed sales — went from $5,300 per week to $10,700 per week. Nothing else had changed. Same Amazon campaigns. Same prices. Same listings.
Statistics and Metrics report on PROPAMP AI — where the seller verified Claude's numbers. Click to expand full image
The agent kept going:
Claude Code:
"Post-test week 2 is the cleanest read so far. The last 14-day attribution window from the test closes before that week starts, which rules out late conversions as the sole driver. tACoS still reads 10.7%, which points to ranking — four weeks of higher sales velocity feeding back into the listing's position. The next few weeks will show how much of the lift sticks."
What comes next
Next on his list: three more tests across his other products — different price points, different influencer pools. Same setup, same metric stack.
One thing he's adding to the methodology: tracking keyword rank in parallel with each campaign. Over four to six weeks post-test, late conversions taper off while a ranking lift holds. With rank history and Statistics and Metrics side by side, he can separate the two effects cleanly in future tests.
If External ACoS reads high again on the next influencer tests, the seller now knows what that number is — direct attribution inside the 14-day window — and what other metrics to check before closing the test.
Setup
Claude Code by Anthropic - $100/month
PROPAMP AI + MCP server- connects the agent to everything inside the platform: live ad performance data PROPAMP AI pulls from Amazon (Amazon Ads campaigns plus Amazon Attribution for off-Amazon clicks and sales) and from Google/Meta ad accounts, plus manual influencer campaigns the seller logs by hand - $47/month
Time spent on the test wrap-up with the agent's analysis included: ~10 minutes.
