Paul walks through a live experiment gone slightly sideways: turning on a winning conversion value setting instantly doubled ad budget, and the algorithm responded by flooding the funnel with junk leads for a full day before quality started rebounding.

The bigger lesson: even when lead quality looks worse on the surface, the actual volume of good leads was climbing the whole time, it just felt like more junk because there was simply more of everything.

The back half digs into a research idea worth testing further: trying to reverse-engineer what Meta's algorithm (Andromeda) can and can't detect about a business's likely budget, using employee count, job title, and demographic signals to steer ad spend toward better-fit buyers instead of hoping the algorithm figures it out on its own.