Popunder Buying Measured Against a Server-Side Holdout

A pop campaign reporting 2.1x in the network dashboard and 0.9x in your warehouse is not a reporting bug. It is usually the same person twice, sold to you by zones that resell your own retargeting pool back at a CPM. So the first thing built here is the instrument that tells those apart: a geo-split holdout in your own tracker, held for fourteen days before daily spend goes past four figures. Everything below runs downstream of that one number. The buy: Accounts open in your name at PropellerAds, Adcash, Clickadu, RichAds and Adsterra, funded from your card, and they remain yours afterwards. Bidding starts at network-average CPM on a whole-country target for three days, purely to harvest zone IDs. From day four the campaign is rebuilt as per-zone bids and the country target is retired. Frequency is capped at one impression per user per 24 hours. Dayparting for Australian offers sits between 17:00 and 23:00 AEDT, because that is when a pop surfaces to a person rather than waiting in a tab until morning. The prelander, and why it is built the way it is: A pop lands behind what the visitor was doing. They see it when they close that, somewhere between forty seconds and forty minutes later, and they give it about two seconds. That budget dictates the build: under 120KB, first paint inlined, TTFB under 400ms from a Sydney and a Singapore edge, no webfont request, no consent modal on non-EU traffic. We write it. If your team would rather own it, you get the HTML and the reason behind each constraint, and we measure whichever version you choose to run. Fraud here, and what it is not: Pop traffic is largely real humans on real machines, so datacentre-ASN filtering and headless detection catch less than they do in display. The loss is laundering: a zone reported as a mid-tier news domain that is in fact an iframe chain terminating on a stream site. We resolve placements from your tracker's referrer chain rather than the network's report, run IPQS or Anura on the click where volume justifies the per-check cost, and treat any zone whose reported domain and observed referrer disagree as dead on the first occurrence rather than the third. Measurement: Voluum, Binom, Keitaro or RedTrack, yours, on your domain, with server-to-server postbacks and conversions written from your backend. We work in whichever of the four you already have. If you have none, Binom on your own VPS is what we stand up, because the log-level data stays with you and per-event pricing stops mattering at pop volumes. What we refuse: Offers that need the landing page cloaked from the network's own review. Sources that cannot return a resolvable placement domain for at least eighty percent of spend, because if we cannot see where the impression happened then the blacklist is superstition. Incentivised and forced-click inventory, whatever the self-serve UI calls it. Who should not buy this: Anything with a ninety-day sales cycle and a dozen closed deals a quarter. The holdout needs conversions to resolve, and under roughly 300 a month you are paying us to have opinions. Industrial and capital-equipment advertisers almost always sit here; we say so on the first call and it costs us the work. Brand-sensitive advertisers too. Pop inventory runs next to content you would not put a media plan against, and no whitelist makes that stop being true. Reporting: Weekly: spend, live zone count, acquisition cost from your tracker and from the network side, and the gap between them. Monthly: the holdout read with its interval, zones added and killed with a reason each, and a judgement on whether this line still deserves its budget. Twice we have written that last item as a recommendation to stop.