Amazon Wasted Spend Analyzer

Problem

Amazon ad budget leaks quietly. Search terms that never convert, keywords with high spend and no sales, competitor ASINs that soak up clicks. Finding them meant exporting the search term, keyword and target reports every week and reading thousands of rows by hand. It took 8 to 10 hours a week and it was the kind of job that gets skipped when the week gets busy.

Approach

I built a skill in Qore, the AI builder inside the Trellis ads platform, that runs every Monday without being asked.

It reads the search term, keyword and target reports together, then sorts the waste into tiers instead of treating it as one bucket. Anything that has spent more than $15 with no sales at all gets flagged. Anything below 1.75 ROAS gets banned. Anything between 1.75 and 2.0 is close enough to break even that it gets a bid reduction instead, rather than being switched off.

That middle tier was the part worth getting right. A single on or off rule kills terms that only need a cheaper bid.

It also reads back its own history of applied changes, so it can show what it actually did rather than only what it recommended.

Result

• 8 to 10 hours a week back, around 450 hours a year
• Runs unattended every Monday across the US and German accounts
• Near breakeven terms get corrected instead of cut, so working spend is not thrown away
• Thresholds are settings, not code, so the rules can be retuned without rebuilding anything