For a long time, soft launch followed a pretty predictable recipe. Pick a handful of low-CPI markets like the Philippines, Vietnam, or Colombia. Push volume through Meta at CPIs cheap enough that you could run real testing without burning a launch budget. Watch D1, D7, and early ARPDAU. Iterate on the product until the numbers held up. Then flip the switch on global launch and expect the patterns to carry through.
That recipe got a lot of games to market for about a decade. It’s been slowly coming apart for the last few years, and by 2026 most teams still running it are getting results they can’t really trust.
The cheap-volume era of soft launch is ending. Teams that get out ahead of it are going to walk into global launch with better evidence than teams still trying to squeeze insight out of markets that stopped producing it.
Here’s what’s actually changed and what to do about it.
A few things happened at roughly the same time.
CPIs in the traditional soft launch markets climbed. Everybody figured out those markets were cheap, everybody piled in, and the auction dynamics caught up. Philippines, Indonesia, Brazil, and Colombia don’t look like the bargains they used to be for a lot of genres. Even when the absolute CPI is still lower than a T1 market, the gap has narrowed enough that the “cheap volume” label barely fits anymore.
Signal quality got worse across the board, too. iOS changes wrecked the reliability of early post-install signal on iOS, and Android has been drifting in the same direction. Running a few thousand installs and reading D1 and D7 off the platform reports is a lot noisier than it was a few years ago. A lot of that noise ends up getting read as meaningful variance by teams still using the old confidence thresholds.
And the representativeness problem is the one people talk about least. The player who installs a game in a soft launch market, through a Meta campaign optimized for cheap installs, often behaves very differently from the T1 player you’re actually trying to learn about. Engagement patterns, monetization tolerance, time spent, preferred mechanics, all of it can look different enough that “it worked in PH” becomes a weak predictor of “it’ll work in US T1.”
None of these problems is brand new. They’ve just compounded to the point where a soft launch run as a cheap volume engine produces data that can’t carry the decisions teams are trying to make with it.
If the point of soft launch isn’t cheap volume anymore, it’s worth going back to what teams are actually trying to validate in the first place.
There are really only a handful of questions worth answering before a global launch. Whether the core loop holds attention past day 3 in a crowded category. Whether the economy stays stable when different player types engage with it. Whether the LTV curve has the shape you need, not just the level. Whether creative is finding people who actually want the game. Whether the product holds up long enough for monetization mechanics to even get a fair shot at working.
Cheap volume in a low-CPI market is a weak way to answer any of those. It tells you people will install, which is barely a real question in 2026. It gives you early retention numbers that might or might not transfer. It gives you some sense of FTUE completion, which matters, but isn’t the hard part of the validation.
The questions that actually matter need different evidence than what cheap volume produces.
Teams running soft launch well in 2026 are doing it differently.
They’re picking markets based on how closely the player base resembles their target launch market. CPI used to be the deciding factor. Now it’s more like a tiebreaker after similarity. Canada, Australia, and the Nordics are getting used more this way, because the player behavior and monetization patterns there are close enough to bigger T1 markets to actually be predictive. CPIs are higher, but usable data is cheaper than unusable data, whatever the install cost looks like on the invoice.
They’re running smaller cohorts for longer. Instead of chasing volume, a clean cohort of a few thousand users watched past day 30, ideally out to day 60 or 90, tells you more than a noisy cohort of 30,000 read at day 7. The LTV curve doesn’t reveal itself in the first week. Teams that rely on early signals keep getting surprised at global launch when the curve they assumed never actually shows up.
They’re separating product validation from UA economics. A soft launch is supposed to be about whether the product works. UA economics at scale are going to look completely different at launch anyway, so trying to answer both questions with the same test tends to produce noisy data on both fronts. The cleaner approach is running the product test first and figuring out UA economics closer to the target market, later.
They’re stress testing the economy with actual cohorts. Most of the real economy problems show up when whales interact with non-payers in the same meta, or when a specific archetype finds a loop that breaks the intended curve. Early ARPDAU averages miss that completely. What catches it is real cohorts playing the game for a while, and analysts looking at the shape of the distribution instead of the averages.
They’re testing creative separately, in the target market. The hooks that win in a soft launch market often don’t hit the same way in a launch market. Teams that are ahead of this are running parallel creative cells in smaller slices of the actual launch geography, even when product validation is happening somewhere else. That way, the day the global campaign flips on, they already know which creative lands on the real audience.
A lot of the old soft launch logic was built on speed. Cheap volume meant you could answer questions fast and keep the whole process to a few months. The modern version usually takes longer, because the questions are harder and the cohorts have to be observed for more time to produce answers you can actually stand on.
Teams that try to compress the modern approach into the old timeline end up right back where they started, reading early noisy signal and making launch calls on top of it. The better move is accepting that meaningful soft launch work takes six to twelve months, and building the rest of the roadmap around that reality. Cutting the cycle short is how studios end up launching globally with confidence numbers that were never real to begin with.
You need people who can read cohort data past the first few weeks without panicking on early noise. You need product analytics that can show you LTV distribution shape, not just averages. You need a creative testing setup that can run cells in markets outside of the main soft launch region, which adds complexity, but it’s worth it for the signal you get back.
You also need leadership willing to defend a longer, more deliberate soft launch against real pressure to go faster. That pressure usually isn’t ill-intentioned. Somebody has a launch window in mind, a competitor is moving, a publisher deadline is looming. Shipping a product that was never properly validated into that window is one of the fastest ways a game can die after launch. A longer soft launch done properly is what protects the rest of the investment behind it.
Cheap volume was a shortcut that worked for a long time. The shortcut closed.
Soft launch in 2026 is more about evidence than installs. Teams running it well are choosing markets for similarity, watching cohorts long enough for the LTV curve to show up, keeping product validation separate from UA economics, and testing creative in the markets they’re actually launching into. Teams still running the old recipe are getting answers they can’t trust, making launch calls on those answers, and wondering why the game doesn’t scale the way the soft launch data seemed to promise.
The goal of soft launch is the same as it always was. Figure out whether the product is going to work before you pour a launch budget into it. What’s different now is the evidence you need to answer that question, and where you can actually get it.
If you’re heading into soft launch, or already in one and second-guessing what the data is telling you, that’s the kind of problem we work on all day.
Let’s look at your market mix, your cohort logic, and your creative testing plan, and figure out whether you’re really on track for a strong global launch.
