Case file · VPN · Tier 1–2
1,400 zones cut to 63, and the CPI followed
$1.90 → $1.15
cost per install
4.1% → 7.8%
trial-to-paid conversion
63
zones surviving from 1,400
The problem
A VPN advertiser had been buying pops broad across the US, Canada, Germany, and Poland for five months. Spend sat near $22,000 a month on two networks, installs arrived at $1.90 each, and the finance sheet said the channel lost money once trial-to-paid conversion was counted: 4.1% of installs became subscribers, against a break-even of 6.2% at that CPI. The previous buyer's answer had been more zones. The account was live on roughly 1,400 of them, most under $30 of lifetime spend, which is testing theater: enough spend to exist, never enough to conclude anything.
The approach
The first move was to stop pretending 1,400 micro-tests were data. We consolidated spend into a structured sweep: every zone got funded to exactly three times the install payout, no more, with S2S postbacks carrying the install and, three days later, the trial conversion flag. Tier-1 CPMs run two to five times Tier-2 rates, so the sweep ran Germany and Poland first, where the same evidence costs less, and used the surviving pattern to shortlist US zones instead of sweeping them blind.
Scoring judged zones on paid conversions, not installs. A zone producing $1.10 installs that never converted to paid was killed as firmly as an obvious bot source; a zone producing $1.60 installs with a 9% trial-to-paid rate earned bid increases. By week six the list was down to 63 zones across both networks, and the bid ladder ran inside the whitelist only, climbing while blended CPI stayed under the $1.30 ceiling the payback model allowed.
The result
Week nine closed with CPI at $1.15, trial-to-paid at 7.8%, and the channel profitable for the first time: comfortably past the 6.2% break-even with margin to fund the next round of testing. Volume dipped 18% during weeks three and four while the junk spend drained out, and the write-up keeps that dip on the record, because a whitelist rebuild that promises no volume trough is describing someone else's account. The 63-zone whitelist shipped to the client as a plain CSV, per the managed-pop rule that the list is the client's asset, not our leverage.