How to test a new traffic source before scaling

Tag before you spend

Attribution can only segment what was labeled, so sub IDs, placements, and click IDs go into the URL from the very first click. There is no retroactive tagging: an untagged first week is a week of spend you can never break down, from a seller you now know nothing about.

Watch the first hours live

Open the realtime view while the first clicks land: it shows who is reading right now, split into humans, bots, and named crawlers. A source that delivers automation usually shows its hand inside the first hundred visits, before any daily report exists.

Respect the volume floor

Quality scores respect the volume floor and leave a source unscored until it has sent enough traffic to judge. Apply the same patience yourself: a bad hour at tiny volume is a coin toss, not a verdict, and sellers deserve the same fairness you want from your own buyers.

The disqualifiers

• A confirmed bot share that grows as volume grows: the seller is topping up the order with automation.
• Uniform sessions: identical duration, identical depth, identical bounce. Real audiences are messy.
• Geography that does not match what you ordered, especially one city carrying a country sized order.
• A desktop heavy mix on an offer that only converts on mobile.

Scale the placement, not the source

Good sources are rarely uniformly good. Scale the placements that survived, blacklist the ones that did not before adding budget, and when several fail at once, the ranking in which sources to cut first applies unchanged. Download the evidence CSV while the bad placements are still sending, because once blocked they stop generating their own proof.