Why Do Most Companies Fail When Learning from Supercell's Small Team Model?

Deep thoughts on AI and aspirations —— ByteDance Deep Thinking Circle

Supercell is one of the most studied companies: founded in 2010, a Finnish company with around 300 employees, generating $1.56 billion in revenue in 2019, with revenue per employee exceeding 36 million RMB. Clash of Clans, Clash Royale, and Brawl Stars all came from here. The “middle platform” concept that was once popular in China’s internet circles also drew inspiration from it.

What people typically learn from it comes down to two words: small teams, decentralization. CEO Ilkka Paananen repeatedly says that decisions should be made by the teams most capable of making them, and his goal is to become “the world’s least powerful CEO.”

But I’ve always had a question: this story has been told for over a decade, so why hasn’t there been a second Supercell?

After reading Paananen’s tenth-anniversary retrospective, my answer is: most people see the decentralization but miss its prerequisites; they see the organizational form but not the expensive decisions behind it.

Three Prerequisites Behind Decentralization, Each One Costly

Let’s look at the prerequisites first. Supercell’s decentralization works because three conditions are simultaneously met.

The first is talent density. Supercell has a hiring principle passed down from its first chairman: when considering whether to hire someone, imagine the company’s current average talent level—only hire if the new person will raise that average. After this kind of screening, what remains are people who can be completely trusted. Decentralization is delegated to this kind of person. Without sufficient talent density, decentralization isn’t trust—it’s abandonment.

The second is cash and patience. Teams have the authority to kill projects, even good ones—both Smash Land and Rush Wars received positive feedback in early testing, yet teams judged they weren’t games that could be played for many years and killed them themselves. This freedom requires an extremely patient capital structure: able to withstand repeated restarts and entire quarters without new product launches. The Clash of Clans team once spent over two years paying down design and technical debt, prioritizing fundamental problems over chasing short-term metrics—something that simply cannot grow in companies driven by quarterly targets.

The third is long-term signals. Supercell’s teams judge everything by the same measure: how many years will players play this? This signal is more stable than revenue and more honest than daily active users. Without such a measure, decentralization loses direction—teams don’t know how to self-check and can only turn back for the boss’s verdict.

These three conditions add up to one sentence: decentralization is a luxury good, and you have to be able to afford the price first.

Culture Is Defined by the Most Expensive Decisions

Now let’s look at this company’s true core. Paananen has a view I strongly agree with: culture is defined by the most difficult decisions, not by slogans on walls or slides.

The most典型 example is Boom Beach. At a critical leadership meeting five months into development, 9 out of 10 people thought the game should be killed. From a business perspective, killing it was rational. But management thought of another layer: if they vetoed what the team wanted to do today, the independent team culture would end there.

They chose to listen to the team. It turned out the team was right this time. But Paananen emphasizes the point that follows: even if the team had been wrong, the company’s decision would still have been right.

This statement is worth pondering because it reveals the mechanism of trust. Trust isn’t a slogan—it’s a commitment that requires real money to honor. Trusting only when you’re confident isn’t called trust, it’s called agreement. Teams dare to take risks, dare to kill their own projects, dare to speak the truth because they’ve witnessed firsthand: the company pays more for its culture than for any single project.

There’s another small detail. Over these years, several team managers at Supercell voluntarily stepped aside, giving their positions to people they considered more suitable—no one required them to do this. Others chose to kill their own projects and moved to help other teams launch games. Such behavior cannot be produced by performance systems; it can only come from a shared understanding: this company is playing an infinite game, not competing for next quarter.

Two Pitfalls When Learning

If you really want to learn, first see clearly the two places where it’s easiest to fail.

The first pitfall: adding rules whenever something goes wrong. This is human instinct, and Supercell stepped into it themselves. They once mandated that all new games must reach a playable stage within 3 months, intending to accelerate validation. The result was that developers secretly started development early to meet the target, spending energy on circumventing rules rather than making good games. This rule was later deleted, returning to the original point of “believing each team will make the right decision.” The mechanism is clear: rules optimize metrics, people then optimize the rules, and whatever you write down becomes a loophole to exploit.

The second pitfall: treating survivorship bias as methodology. Paananen himself says at the beginning of his retrospective that their success has a large element of luck—don’t learn Supercell’s culture, build a culture that fits your own company. This isn’t modesty. Behind every Supercell that survived, there are numerous companies that also tried small teams and decentralization but quietly died—no one writes their retrospectives. Decentralization without talent density and cash backing is chaos; killing projects without long-term signals as a basis is willfulness.

One more observation relevant to today. Stories of small teams in the AI era are becoming more common, and those three prerequisites are actually being quietly rewritten by technology: AI has taken over much execution work, tools have replaced part of coordination costs, and companies that couldn’t afford or delegate before now have some learning space. But one thing cannot be replaced—the ability to make expensive decisions. Whether to kill a project, whether to listen to the team, whether to withstand this quarter—in the end, it’s still human judgment.

Before Learning, Ask Yourself Three Questions

If you want to try small teams and decentralization in your own company, do three self-checks first.

First, ask about talent: Can your direct reports operate normally without your approval? If not, solve hiring first, then talk about decentralization—reversing the order is abdication.

Second, ask about cash and incentives: Can the company bear the team being wrong once? If one failure causes performance evaluations to collapse, no one will dare take responsibility for decisions, and decentralization is just for show.

Third, ask about signals: Does the team have a metric that everyone acknowledges and can look at for ten years? With only quarterly numbers, teams will naturally only look at quarterly numbers.

Only after passing all three checkpoints should you talk about decentralization; whichever checkpoint you fail, fix that one first. Supercell’s story is compelling, but its real value isn’t in providing a template—it’s in clearly marking the price of that template.