Soft launch to global in 11 weeks: the post-mortem
A four-person Munich studio arrived with a finished match-and-sort puzzle game, a €60,000 launch budget, and one prior launch that had died quietly. Eleven weeks later the game went global with an iOS CPI of $1.70, down from $2.90 in week two, and a day-90 cohort that had already paid for itself. Questions by Ozan, answers by Micha and Tereza, exactly as we run internal post-mortems.
$2.90 → $1.70
iOS CPI, week 2 → week 10
29% → 34%
D1 retention against a 31.85% puzzle median
19
hooks tested and killed
Week 11
global launch, gates green
Q: Why did the last launch fail, and what changed this time?
The previous title launched globally on day one with two ads, both made the week before. Nobody knew if the game retained because nobody had gated anything: spend went out, installs came in, revenue didn't. This time we wrote the gates before the first euro moved: day-1 retention at or above 32%, an iOS CPI ceiling of $2.00 at soft-launch scale, and an early revenue slope the payback model could live with. The puzzle category's published medians (D1 31.85%, D7 12.18%) gave us an honest bar instead of a hopeful one.
Q: How did you pick the soft-launch geos?
A two-rung ladder. The Philippines gave cost-efficient Android volume for stability testing and early funnel reads; the Netherlands gave expensive, high-quality signal on iOS monetization, close enough to launch-market behavior to trust. One cheap geo alone flatters you; one expensive geo alone starves the sample. The pairing cost about 20% more than a single-market soft launch and saved the study from the classic error of reading Manila retention as Munich revenue.
Q: Nineteen hooks died. What were the notable funerals?
The prettiest death was a cinematic intro cut from actual gameplay, which everyone loved and nobody installed from: 1.9 installs per mille against a bench median of 3.4. The most instructive was a timer-pressure hook that pulled excellent CTR and terrible players; its cohort's D1 came in seven points under control, which is how a creative fails after the click. Both have headstones in the Graveyard. The winner, hook 17-A, showed a stranger sorting the puzzle wrong and let the viewer's correction instinct do the rest.
Q: Where did the CPI drop actually come from?
Three places, in decreasing order. First, the hook itself: 17-A's install rate roughly doubled the launch control, and in an auction, install rate is a discount. Second, concentration: once the winner was proven we stopped spreading budget across polite mediocrities. Third, platform mix: Android carried volume while iOS scaled carefully, since global averages still run near $1.20 Android against $3.60 iOS and the game monetized well enough on both to respect that split. Nothing clever, applied weekly.
Q: What almost went wrong?
Week seven. CPI had touched $1.85 and the studio, reasonably, wanted the global button. The retention gate said wait: D1 sat at 31%, one point under the gate, following an onboarding change that had quietly lengthened the tutorial. Reverting it took four days; D1 recovered to 34%. Pressing early would have meant scaling spend onto a leak. The most valuable thing in the whole engagement might be that four-day pause, which is a strange thing to invoice for.
Q: What does "profitable D90 cohort" mean here, precisely?
The week-2 through week-5 soft-launch cohorts recovered their blended acquisition cost from in-app purchases and ads revenue inside 90 days, measured in the MMP with SKAdNetwork postbacks filling the iOS gaps. It does not mean the whole spend was profitable by launch; it means the unit economics were proven before scale, which is the only thing a soft launch owes you.
Asked after this story
Not at eleven weeks. Mid-core CPIs run $4.50 to $6.00 on iOS against a puzzle title's lighter costs, and payback curves stretch months longer, so the soft launch needs more spend and more patience per read. The gate logic transfers unchanged; the calendar roughly doubles. Rushing a mid-core soft launch is how studios discover their LTV model was fiction.
Because cheap installs give cheap signals. A single low-CPI market tells you how the game performs among users you won't be buying at global prices. Pairing one signal-quality geo with one cost-efficient geo let the studio read monetization honestly while still affording volume, which is the entire trick of the ladder.
The bench kept running, at lower intensity: eight to twelve concepts a month instead of twenty. Fatigue does not respect launch parties. The choice-hook family carried spend for roughly nine more weeks before installs per mille slid, and its successor came from the same test bench, already validated, before anyone panicked.
Planning a soft launch?
The geo-ladder guide publishes the method. The working session applies it to your game.
Start a run