The Ninety-Pull Machine: Inside Genshin Impact's Revenue Architecture and the Esports Gap

At 1:47 in the morning, on my desk in Incheon, there was a file labelled esports. I opened it. No teams. No tournaments. No players. No balance patch of any kind. Only a sequence of numbers repeating like the ticking of a clock set to the wrong time zone: 90. Ninety pulls to be certain of touching a five-star character. Split into two phases, roughly 21 days each, each phase with its own banner.

I read that file three times, not because the content was difficult, but because I wanted to be sure I had not misread the label. Nine years covering this industry, from the sleepless nights at Incheon Samsan watching Longzhu beat SKT T1, to the data analyses I dashed off at 2 a.m. for Inven Global, I had never seen a document so badly mislabelled. And I had never seen a document, however mislabelled, lay bare so clearly the machinery running behind most of the game industry's revenue today.

Ninety is not a competition rule. It is a price.

The original content, once deconstructed, contains 28 information points. They describe the Genshin Impact banner schedule: phase two of version 7.0 brought rerun banners for Flins and Ineffa; version 7.1 opened with two new characters at once, Vesna and Vodyanitsa, then phase two returned with Skirk and Escoffier. A few other names appear scattered throughout: Odette, Aino, Iansan, Lan Yan. No rankings, no win rates, no rosters.

The mechanics section is far clearer. The pity system guarantees a five-star character within 90 pulls. The 50/50 rule states that the first five-star has a 50% chance of being the featured character and a 50% chance of being a standard character; if it misses, the next five-star is guaranteed to be featured. Pity is shared across banners of the same category. And a separate banner lane called Chronicled Wish exists for older characters. The rerun schedule is not fixed: some characters are absent for more than a year, others return within a few versions.

One detail made me pause longer than anything else: of those 28 information points, 20 carry no source. Exactly one cites an official publisher announcement. Three are the author's subjective opinion. The article itself concedes that the exact banner schedule is still to be confirmed.

So I was holding a document describing a revenue machine, built on unverified facts, and mislabelled by industry.

Start with the price. The 90-pull guarantee is a price ceiling designed never to be called a price ceiling. It creates a feeling of accessibility: players know exactly how much they need to be certain of a result, so they no longer feel they are throwing money into a bottomless pit. But that ceiling only holds for one five-star, and only holds if players accept that this five-star may not be the character they wanted.

That is where the 50/50 mechanic enters. If every five-star landed exactly at the 90 threshold, the expected arithmetic for a featured character would be: a 50% chance of needing 90 pulls, a 50% chance of needing 180 pulls, an average of 135. Players remember 90. They pay 135. And 135 never appears on any promotional banner.

I write this not to indict a mechanic. I write it because it is a design lesson the esports world has missed for years: what gets communicated is the floor, what gets paid is the expected value. In esports, the floor is a group-stage slot, the expected value is the trophy. Nobody sells fans a group-stage slot and calls it a cup.

People look at drop rates. I look at the cracks in the revenue architecture.

The shared-pity mechanic carves another crack. When pity accumulates across banners of the same category, the marginal cost of switching from one banner to another is close to zero. A player who has banked 60 pulls on a previous banner feels nothing is lost by trying their luck on the next one, because that progress follows them. Technically, this is a decision that lowers friction. Economically, it is a decision that raises spending frequency. Low friction and high frequency are two sides of the same revenue equation.

The unfixed rerun schedule carves the deepest crack. No advance publication. No cycle for players to plan their finances around. A character can be absent for more than a year and then return in silence. That uncertainty is designed. Whoever does not know what tomorrow holds will always prepare more than whoever does.

Then comes Chronicled Wish. I read this information point and thought immediately of a familiar concept in sport: the legends match, where retired names are invited back onto the pitch one more time. Chronicled Wish does exactly that with characters. It opens a separate lane for names that have gone quiet, so they do not crowd the main banner rhythm, yet still generate revenue. This is a second monetisation tier, and it is clever in that it does not disrupt the rhythm of the first.

Transfers are not transactions, they are unfinished love affairs stitched back together. Rerun banners work the same way. They resell an old love to those who missed it, and sell again to those who already have it but want more.

The final element in this architecture is packing two new characters into a single phase. When Vesna and Vodyanitsa appear together in phase one of version 7.1, resource-allocation pressure peaks right there, not because both are strong, but because both are new, and nobody wants to miss new. This is where I see the difference between the two models most clearly: esports monetises by stretching attention across a season, this model monetises by compressing attention into a 21-day window.

The esports model I follow every week depends on three parties: sponsors, broadcasters or streaming platforms, and the publisher sharing revenue. Three parties, three calendars, three risks. This machine has one party. The publisher sets the rules, publishes the announcements, collects the money, and is also the only source of the official announcement that article cites. That concentration of power makes this value chain resilient to the calendar shocks esports cannot withstand, but it also makes it wholly dependent on a different variable: regulation.

I once sat and rewatched 12 DAMWON matches to write a series about Canyon, and what I learned then was that data only means something when placed in a causal chain. The chain here is: the floor creates a sense of safety, the 50/50 creates variance, shared pity creates frequency, an unfixed schedule creates contingency, a double banner creates the peak. Five steps, closed loop, not a single match required.

There is a very comfortable temptation: call this machine gambling and end the article there. I have seen hundreds of such pieces, and most of them are lazy in exactly the way they accuse others of being. Gacha sits outside the legal definition of gambling in most current markets, because it is a randomised reward mechanic with a guaranteed floor, and that guaranteed floor is precisely what separates it from a roulette wheel. Calling it gambling skips the single most important detail.

But there is a second temptation, more dangerous: call it esports and push it into the same frame of reference as things it shares nothing with. No teams, no players, no tournaments, no balance patches. Labelling this document esports does not make it esports. It only corrupts every analysis that follows.

And this is where I want to linger longest. Of the 28 information points, 20 have no source. Many names and version numbers cannot be cross-checked against any official announcement I could find. Which means the hype wave that piece generated rests on a foundation thinner than the 50/50 mechanic itself. If esports has a lesson to learn here, it is the lesson of source discipline, something I had to forge from the night I interviewed a 19-year-old player in a PC bang and kept the information sealed until exactly 2 a.m. before publishing.

People look at drop rates. I look at source rates.

Every patch is a draft, and only real writers dare to keep writing. But before writing on, perhaps one simple question is worth asking: who winds that 21-day clock? If the answer is the publisher, then every piece of advice to save for the next version is part of the design, not a piece of advice standing outside it. And if esports is to learn anything from this machine, it is this: fans deserve a floor too, a promise that after however many years of waiting, they will be certain to receive something.