Product Ad Underperformer Detector

Problem

Ad performance was managed at the campaign and keyword level. Reviewing individual advertised products meant going through the catalog one product at a time, so it was rarely done. Money was being lost at a level nobody was looking at.

Approach

A second skill in Qore scores every advertised product against a return on ad spend target over a rolling 30 day window, estimates how much of its spend was wasted, and queues the worst offenders for pausing.

The first version paused a product after one bad month. That turned out to be too aggressive, so I added a second test: a product now has to underperform in the previous 30 day window as well before it is recommended for pausing. Products without enough spend or clicks to judge fairly are held back and reviewed over 90 days instead, rather than being punished for thin data.

Nothing switches off on its own. Pauses are queued for approval.

Result

• The two period rule cut one week’s pause list from 126 products to 6
• An audit of roughly 150 products paused under the original single period rule found 36 of them, about $2,000 of spend, were profitable at the moment they were switched off
• Around 3 to 4 hours a week of product level ad review that was previously not happening at all
• Every pause is reviewed before it goes live