Case Study 2: Supercell's Kill-Rate Culture — How Clash of Clans Survived Fourteen Abandoned Projects
In 2012, a small Helsinki studio called Supercell shipped a mobile game called Clash of Clans. Within two years, it was generating over five million dollars per day. By 2025, it had grossed more than ten billion dollars cumulative, becoming one of the most financially successful games ever made. Brawl Stars, Hay Day, Boom Beach, and Clash Royale — all from Supercell — together added several billion more.
What is less widely known: in the same period that produced these blockbusters, Supercell killed — by their own count, with a public ceremony each time — at least fourteen other projects in soft launch or late prototype. Among the announced kills: Battle Buddies (2014), Smash Land (2015), Spooky Pop (2015), Rush Wars (2019), Hay Day Pop (2020), Everdale (2022), Floodrush (2023). Many more were killed before reaching the soft-launch stage and never received public names. The studio's stated kill rate is somewhere around 90 percent — for every project that reaches release, roughly nine others are buried.
The kill rate is not an accident or a sign of weakness. It is the central feature of Supercell's design culture, and it is enabled by a playtest practice — specifically, an analytics-driven soft-launch protocol — that treats every project as provisional until the data says otherwise. This case study examines how Supercell's playtest-and-kill culture works, what it costs, and what it means for studios that cannot afford to operate at Supercell's scale but can borrow the underlying disciplines.
The Soft Launch as Industrial Playtest
Supercell's central institutional practice is the soft launch. A project that reaches a sufficient state of completion is released, in beta or 1.0 form, to a small set of countries — historically Canada, Australia, Norway, sometimes the Philippines, the Netherlands, or New Zealand. These markets are chosen because they are English-speaking (or close to it), small enough that a flop will not damage Supercell's brand globally, and demographically reasonable proxies for larger English-speaking markets. The game is fully playable. Players can pay real money. The build is, to the player, a launched game.
What it is, to Supercell, is a playtest at population scale. The data flowing back — D1, D7, D30 retention; revenue per user; session length; level completion rates; churn points; engagement curves — gets evaluated against thresholds that Supercell has refined across years of soft-launching games. The thresholds are confidential but the principles are visible: D7 retention below a certain percentage means the game is not going to grow virally. ARPDAU (average revenue per daily active user) below a certain dollar figure means the unit economics will not support the studio's overhead. Hours-per-active-user below a certain figure means the game is not sticky.
If the numbers clear the thresholds across a sustained soft-launch period (typically three to six months, sometimes longer), the project is greenlit for global release. If they do not, the project is killed. The kill is not a failure of the team or a black mark on careers; the kill is the system working.
This is, in form, the analytics-based playtest from the chapter, run at industrial scale. Where an indie team might collect telemetry from a hundred testers in a Discord beta, Supercell collects it from hundreds of thousands of paying users across multiple regional markets. The questions are the same: are people retained, are they engaged, are they spending, do they tell their friends. The answers are the same: numerical, ruthless, hard to argue with.
The Public Funeral
What is unusual about Supercell — what makes it worth a case study rather than a footnote — is the cultural ritual surrounding the kill.
When a project is killed, it is announced. Often by Ilkka Paananen, the CEO, in a public blog post or industry interview. Champagne is opened at the office. The team that built the killed game is celebrated for shipping a clean, bold, well-executed project that the data did not support. The team is reassigned, often to seed new prototypes, sometimes to support the projects that did succeed.
The ritual is intentional. Paananen has written and spoken about it at length — the practice is borrowed from Supercell's understanding of how research-and-development cultures work in other industries, where "failed" projects are treated as the cost of doing exploratory work rather than as professional disasters. The champagne is not ironic. It is a deliberate cultural mechanism for keeping the team willing to ship projects that might be killed.
This matters more than it sounds. A studio that says it tolerates failed projects but behaves as though killed projects damage careers will see its teams quietly steer toward safer, less ambitious projects — and the safe projects, ironically, are also the ones least likely to break out into blockbuster territory. The whole system depends on the team genuinely believing that proposing a bold concept that might get killed is a good career move, not a bad one. The champagne is the expensive proof that the studio means it.
Without that cultural commitment, the analytics-driven kill becomes a different beast entirely. It becomes a sword hanging over teams, a constant threat, a source of anxiety that suppresses the willingness to take risks. The data is the same; the culture changes its meaning. Supercell's specific contribution is to have built a playtest practice that the team genuinely accepts as a fair arbiter, not as an executioner.
What Gets Killed and Why
Some specific examples illustrate what soft-launch data reveals.
Battle Buddies (2014) was a tactical multiplayer shooter, soft-launched in Canada. The build was polished, the team was strong, and the early reception was warm. The data, however, showed flat retention. Players liked the game, played a session or two, and did not come back. The hypothesis — never publicly confirmed but implied in subsequent talks — was that the game lacked a meta-progression system that would pull players back day after day. The killing decision was made within months. Subsequent Supercell games (Brawl Stars especially) leaned heavily into meta-progression, applying the lesson directly.
Smash Land (2015) was a casual card-collection game, mash-up between Peggle-style physics and gacha mechanics. Soft-launched in several markets, the game showed acceptable retention but weak monetization — players engaged but did not spend. The unit economics did not work, and the game was killed.
Spooky Pop (2015) was a match-three puzzle game in the Candy Crush genre. Soft-launched, it generated moderate numbers — not bad, not great, somewhere in the "okay" range. Supercell killed it anyway. The reasoning, articulated in a Paananen interview later, was that "okay" was not enough. A game that would clear a lower studio's release threshold did not clear Supercell's, because Supercell's overhead, ambition, and team allocation were structured to make blockbusters or nothing. The killing decision was about opportunity cost: the team and resources tied up in Spooky Pop could be applied to a project with breakout potential, and shipping a moderate game would consume both the resources and the team's morale for a return that would not justify either.
Rush Wars (2019) was a base-builder with strong early reception in soft launch. The numbers were promising. The team was experienced. Then, after several months of polish and additional soft-launch markets, the data plateaued. Retention was good but not great. The decision to kill — visible in industry coverage at the time as a surprising one — was again about thresholds. Good was not great, and Supercell ships only great.
A pattern: the kill decisions are not about "the game is broken." They are about "the game is good but does not clear the bar required to justify the studio's resources, ongoing operation, and team morale." This is a much higher bar than most studios apply, and it is the bar that Supercell's soft-launch playtest data is calibrated to evaluate.
The Discipline of Ignoring "Friends Like It"
A common indie failure mode, discussed in the chapter, is over-weighting feedback from friends and family. Supercell's playtest practice is structurally immune to this failure because the population scale dwarfs any social connection the team might have to the players. When you have 200,000 soft-launch users in Canada, your roommate's opinion is not visible in the data. The team's belief that the game is good — and they almost always believe it is good, because they would not have spent two years building it otherwise — is similarly invisible. The numbers are the numbers. The team's emotional investment cannot influence them.
This produces the conditions under which a team can sincerely build, ship, watch fail, and accept the kill, without the failure being personal. The data is impersonal. The decision follows the data. The team did not fail; the data revealed that the design hypothesis did not pay off at scale, and that is information.
For studios that cannot operate at Supercell's scale — which is to say, almost every studio — the lesson is not "do exactly what Supercell does." The lesson is separate the personal investment from the data. Build mechanisms by which the data is visible to everyone, evaluated against pre-committed thresholds, and used as the actual basis for go/no-go decisions. The thresholds matter. Without pre-committed thresholds, the data becomes another input that the team rationalizes around; with them, the data is binding.
The Cost of the Kill Rate
Kill rates are not free. Supercell employs hundreds of people. Each killed project represents months or years of work by a team of fifteen to thirty. The gross sunk cost of the killed projects across Supercell's history is in the tens of millions of dollars, possibly more.
The math works because the surviving projects are enormously successful. Clash of Clans alone has earned over ten billion dollars. The math does not work for studios whose blockbusters earn a thousand times less. A solo developer cannot afford to kill nine projects to ship one; the survival math requires the breakouts to be capable of returning many multiples of the kill cost.
This is a real constraint. The Supercell model is not directly transferable to indie scale. What is transferable is the underlying discipline: the willingness to evaluate projects on data rather than on hope, the pre-commitment to thresholds, the cultural acceptance that a project that does not clear its threshold should not ship. At indie scale, this might mean killing a prototype after three weeks rather than persisting for two years. It might mean a soft launch on itch.io with a free build, watching whether players actually return, and using the data to decide whether to invest in the polish phase. The shape of the practice is recognizable; the scale is different.
What to Take From Supercell's Practice
Five takeaways an indie or mid-sized team can adapt:
- Pre-commit to thresholds, in writing, before you have data. Decide what numbers would justify continuing, what numbers would justify killing, and write them down before you have any results. The thresholds bind future-you in ways that present-you cannot wriggle out of.
- Soft launch is a playtest at scale. The Supercell ritual of releasing in small markets is, structurally, an analytics-driven population playtest. A free itch.io release, a Discord-only beta, or a Steam Early Access cohort can perform the same function at indie scale. Use the data the soft launch generates the way Supercell does — as binding evidence about the game's prospects, not as a polite suggestion.
- Celebrate killed projects in public. This sounds insufferable until you watch a team that does it. The reason it works is that it makes proposing risky projects safe, which is what you need if you want to occasionally land a breakout. Without the public celebration, the team learns to play safe.
- Separate emotional investment from the data. Build mechanisms — review meetings, public dashboards, pre-committed thresholds — that make the data visible regardless of how the team feels about the project. The team will always feel the project is good. The data is what tells you whether the players agree.
- Accept the cost of kills as the price of breakouts. Studios that ship every project they start will tend to ship many mediocre projects and few great ones. Studios that kill aggressively will ship fewer projects, more of them great. The math depends on what your studio is trying to be.
The Two Cases, Together
Valve and Supercell represent two distinct expressions of the same underlying conviction: that the only reliable signal about a game is the data from real players, and that the practice of designers and studios must be organized around capturing and acting on that signal. Valve operationalized it through small-scale qualitative observation — the cabal, the watching wall, the camera on the face — applied with discipline across many design decisions over long development cycles. Supercell operationalized it through large-scale quantitative observation — the soft launch, the threshold metrics, the public kill — applied with discipline across the project-greenlight decision itself.
Most studios operate with neither model in pure form, and most studios ship games that fail in ways the model in question would have caught. The instructive thing about both Valve and Supercell is not that their methodologies are uniquely insightful — the underlying ideas are old, the practices are publicly documented, the tools are available. The instructive thing is that both studios built their cultures around the conviction that playtest data is binding, and the cultural commitment is what makes the methodologies work.
If you take only one thing from this case study and the previous one, take this: the playtest practice you can install in your team is exactly as effective as the cultural commitment your team has to acting on what the practice reveals. Without the commitment, the practice becomes theater. With the commitment, the practice becomes the engine that produces games people actually want to play.
Supercell ships one game for every nine they kill, and the one is Clash of Clans. Valve takes seven years to ship a game and the game is Half-Life 2. The arithmetic is different. The conviction is the same.