Unpredictable Hits #013: A Cat Sudoku Doing $200K a Day + Roblox Hands an Engine to 132 Million People
From the Editor
What's up friend?
First of all, go watch The Odyssey if you haven’t. Yeah it is as good as people claim. It’s also three hours long and the medium-sized popcorn is not enough. You should buy the large one, learn from my mistakes or bear the consequences.
Now I feel I contributed to your happiness and we can start this edition.
For 18 months, a game concept sat in the open on Kwalee's portfolio. Minesweeper crossed with Sudoku. Anyone in the industry could have picked it up. Nobody did.
Then a Chinese studio wrapped it in cats, pointed ten thousand AI-generated creatives at it, and within 60 days Meowdoku was at 2 million daily active users and an estimated $200K a day in pure ad revenue. The idea was free the whole time. The execution was not.
That story runs through this entire issue.
What else?
Amit Carmi sat down with Plarium's Matt Veysberg, who runs a 50-plus-person creative org and still insists you only need one winning creative.
We published our teardown of how Grand Games launched Block Out with over 1,000 creatives after going nearly a year without greenlighting a single game.
618 hybrid-casual games launched in Q1 and four of them matter.
Frank Gibeau, Zynga CEO, explains how it went from a $1.87 stock to a $12.7 billion exit and shares a ton of game wisdom. Last time I had the privilege of meeting Frank he spoke about the days he worked on the first ‘The Sims’ in EA - highly recommend you spend some time with this one.
Roblox just handed a game engine to 132 million people. And the most honest AI story in gaming is about a team that shrank from 25 people to 3.
Let's go.
Sett Stuff and Analyses
Unpredictable Hits, Season 3: Matt Veysberg on the One Winning Creative
Season 3 opens with Matt Veysberg, Director of Global Creative Marketing at Plarium, the studio behind RAID: Shadow Legends. Fifteen years of making the case for judgment over volume, now running a 50-plus-person creative org. Amit hosts.
The claim: The networks push everyone to scale from 20 creatives to 1,000. The biggest live titles run the same winner for months. In Matt's words: “The thing is that you only need one.” Volume feels productive. Finding and owning a winner is what pays.
The taste problem: “The algorithm doesn't have a taste.” Meta distributes impressions. It has no opinion on your joke. The edge is methodology: creative pillars tied to real motivations, a reason behind every test, and the discipline to read the results.
The missing hire: Product marketing done right is a marketing lead embedded in the product, the person who turns deep mechanics into motivation and messaging. Most studios still don't hire for it.
On brand: Brand runs on a two-to-three-year clock, separate from performance marketing. Great brand makes you feel something, the way RAID's cinematics or the Coca-Cola Santa do. “If it doesn't make you feel anything, this is not brand, man.”
On AI: Soon a schoolgirl will make a high-end cinematic trailer. When the tools are free, the edge is taste, experience, and knowing your audience. AI lets a team of 30 hit like 300. The human still decides what's good.
Grand Games Spotlight: Sniper on the Game, Shotgun on the Ads
We published our teardown of the Block Out launch, and it holds the whole Grand Games playbook in one contradiction. Brutal patience on which games get made. Total aggression on the ads once one does.
The sniper: Grand Games went close to a year without greenlighting a single new game. Every candidate has to show real differentiation in art, level design, or core product, and survive a check against Sensor Tower market data before anyone prototypes.
The shotgun: When Block Out launched, it launched with over 1,000 creatives. 260 playables, 888 videos. Their previous launch went out with something like 20. That is more than fifty times the creative volume, on a sort puzzler scaling at an estimated $300K a day on iPhone alone. The figures come from UA-community teardowns, so treat them as estimates. The shape is what matters.
The company behind it: Two years old, six games shipped, a $500 million annual run rate, and still only about 75 people. Five autonomous studios of roughly fifteen, each shipping without asking a founder.
The lesson: Block Out itself is Proven and Better, almost nothing New, to borrow Mark Pincus's greenlight framework. The New moved into the creative. You can't predict a hit. You build the machine that finds it.
Point a machine at your next idea.
Meowdoku ran ten thousand AI creatives at an idea anyone could see. Sett is that machine, for your studio.
Industry stuff worth knowing
1. 618 Hybrid-Casual Games Launched in Q1. Four Matter.
Author: Matthew Emery, Gamigion (data: AppMagic)
Last issue we counted Q1's casual launches: 262 games, three winners. Hybrid-casual makes that cohort look roomy. 618 games launched in Q1, and after three months of baking, four are running at $10M+ a year.
The tiers: 0.5% of launches (3 games) hit $20M+. 0.6% (4) hit $10M+. 0.8% (5) hit $5M+. And 9.1% (56 games) cleared $1M+ annualized. The prior quarter's $10M rate was 0.7%. As Emery puts it: “Low, and remarkably consistent.”
The winners: Amaze GO! by Learnings ($33M, arrow mechanic), Jewel Coloring by TaTa Game ($28M, sort), then Bus Traffic Fever! ($21M) and CubeAway ($11M), both from GOODROID, Japan's top hybrid-casual publisher with 370 releases since 2015. The top two run zero IAP. Pure ad revenue, no ad-removal option, nothing.
The mechanic shortlist: All four winners run one of two mechanics, arrow or sort. Outside those, the $10M tier was empty. Emery's line: "Winning mechanics: arrows and sort."
Emery's kicker: None of the top four win on game quality. They win on portfolio-level UA efficiency. The game is the ad inventory. The machine around it is the business.
2. Meowdoku: 2M DAU in 60 Days, $200K a Day, Zero IAP
Authors: Felix Braberg and Matej Lancaric, Gamigion (two pieces)
The cold open of this issue, in full. Meowdoku is Minesweeper crossed with Sudoku, wrapped in cats, built by Oakever (real name Learnings / Beijing Lexing). Yes, the same Learnings that tops the hybrid-casual chart above with Amaze GO!. The concept sat unused on Kwalee's portfolio for 18 months before Oakever executed on it.
The numbers: 2 million DAU in under 60 days. Roughly 400,000 downloads a day. An estimated $200K a day in revenue, 100% from ads, against an estimated $500K a day in UA spend. That gap is a bet that the retention curves pay it back.
The creative machine: 10,187 new creatives in the last 30 days, per Braberg. Lancaric puts AI production at near-100%: Impulse-style brain-training hooks, TikTok-comment-style overlays, localized across every language. This is what creative volume as a weapon looks like.
The monetization screw: An ad after every level, 80%+ interstitials, and a retry cooldown that tightens as you get hooked. 120 seconds in your first two days, 100 seconds through day seven, 90 seconds after that. Engaged players tolerate more ads, so the game serves them more ads.
The company: Fully bootstrapped. 70M MAU across the portfolio, a billion cumulative downloads, five straight years of 100%+ revenue growth, and net margins north of 85%, per Lancaric.
3. Your Game Has 47 Seconds to Shine
Author: Lexi Sydow, Chief Analyst & Founder of The ATTN Economy, on 80 Level
The average attention span on a single screen is down to 47 seconds, from 2.5 minutes in 2004. Sydow's argument: your game competes with every app, feed, and platform on the phone, and the five biggest companies in the attention economy are worth a combined $15 trillion. Attention is the scarce resource. Treat it that way.
The case study: Brawl Stars partnered with Taika Waititi. 23M+ views across TikTok, Instagram and YouTube in three weeks, DAU up 25% in March, player spending up 30%, an extra $9.7M month-over-month to $42M in net-to-publisher revenue.
The format shift: Per Sensor Tower's State of Gaming 2026, video is over 53% of ad creative, and playables have doubled to nearly 13%. Attention goes to formats that demand participation.
The thesis: “The studios winning long-term aren't just running events and optimized UA programs. They're building identity.” Paid traffic is transactional. Earned attention compounds.
4. Frank Gibeau on the Zynga Turnaround: $1.87 to $12.7 Billion
Source: Gaming Founders Podcast (Eden Chen, Kevin Zhang)
Frank Gibeau arrived at Zynga to find a $1.87 stock, $600 million in revenue producing $17 million in profit, and an enterprise value around $200 million. His diagnosis: “This is a very fast race car, but the parts are all over the garage floor.” An hour of Gibeau naming numbers most execs won't say out loud.
The surgery: He replaced 16 of 17 VPs and cut 120 active projects down to about six. Then he sold the headquarters building at the peak and spent the money buying studios: Gram Games, Small Giant, Rollic, and Peak, a courtship that started at a GDC party years before the deal closed.
The exit: Take-Two bought Zynga for $12.7 billion. It was the largest deal in interactive entertainment history for about 10 days, until Microsoft announced Activision.
Why UA people should care: Console's capital intensity sits in development, nine figures before launch. Mobile's sits in performance marketing. “For a very small amount of money I can figure out if Match Factory's a hit or not, and then scale it.” And his creative volume number: game ads, “you're doing like 60,000 of them for launches sometimes.”
The AI part: Two-thirds of Zynga uses AI daily and prototyping output is up around 10x. When CEOs announce they'll cut 20% of jobs with AI, Gibeau sees a company demotivating itself at scale. His verdict on that: “Just… well done.”
5. Roblox Hands a Game Engine to 132 Million People
Source: Roblox Newsroom (Nick Tornow, SVP of Engine, and Vlad Loktev, Chief Creator Ecosystem Officer)
Roblox announced Build, a mobile-first AI creation tab that turns a text prompt into a basic playable game inside the Roblox app. Public alpha starts July 28 in New Zealand, age-gated 9+ for testing and 16+ for globally published games. Free base tier, paid options for power users.
The scale: Roblox frames its entire 132 million daily active users as the addressable creator base. Alongside Build come AI agents for playtesting, analytics, and experimentation aimed at professional creators.
The slop question: Roblox answered it before anyone asked, in its own announcement copy: “Our discovery systems are designed to highlight games with long-term retention, which doesn't include AI slop. The quality of games on the homepage isn't changing: If no one plays it—no one can find it.”
The read: When 132 million people can prompt a game into existence, the bottleneck moves from creation to discovery. The moat is retention, and Roblox just said it out loud.
6. The Most Honest AI Adoption Story in Gaming
Author: Michail Katkoff, Deconstructor of Fun
The press version: InnoGames used AI to keep Sunrise Village alive. The actual version, from Thomas Lehr, the InnoGames Director of Engineering who built the system: AI shrank a 25-person team to 3, the other 22 moved to a new title, and the game kept declining anyway. Downloads halved. Revenue fell by a fifth. But the title went from deep red to green.
The real unlock: One absurd manual bottleneck, design data shuffled between Confluence, Excel, CSV, and Unity, fixed with a custom tool built in about an hour. That discovery beat everything the models did. Watch the workflow before you buy the AI.
The unfashionable choice: InnoGames ran GPT-4o for a year and a half rather than chasing newer models, switching only when deprecation forced it. As Katkoff frames it, stability beats capability the moment a workflow runs without a human checking every step.
The honest parts: Lehr admits real skill erosion on the team and real AI fatigue: more output, more decisions, more context-switching, people ending the day more exhausted. His one on-record line about the revenue curve: “At least it slowed the decline drastically.”
Katkoff's verdict: “AI bought the game a longer, cheaper death, and freed up 22 people to go build something with upside.” Against a GamesIndustry.biz survey where 84.2% of professionals say no amount of AI-generated content is acceptable at any stage of development, this is what adoption actually looks like.
👉 Read on Deconstructor of Fun
Meowdoku FAQ
Is Meowdoku made with AI?
The game itself is two old mechanics and a cat: Minesweeper crossed with Sudoku, built by Oakever (real name Learnings / Beijing Lexing). The AI sits in the marketing. Matej Lancaric puts creative production at near-100% AI, and Felix Braberg counted 10,187 new creatives in a single 30-day window.
How does Meowdoku make money?
Entirely from ads, with zero IAP. Estimated revenue runs about $200K a day against an estimated $500K a day in UA spend. The monetization screw is an ad after every level, 80%+ interstitials, and a retry cooldown that tightens from 120 seconds in your first two days to 100 seconds through day seven and 90 seconds after that.
How many people play Meowdoku?
2 million daily active users, reached in under 60 days, with roughly 400,000 downloads a day. The publisher behind it carries 70M MAU across its wider portfolio and a billion cumulative downloads, all fully bootstrapped.
When did Meowdoku come out?
Meowdoku reached 2 million DAU in under 60 days, which puts its launch roughly two months before this issue. The idea behind it is far older. The concept sat unused on Kwalee's portfolio for 18 months, in the open, before Oakever picked it up and executed on it.
Wrap Up
Something kept bugging me while I put this issue together. "Ideas are cheap, execution is everything" is the oldest cliché in gaming. Every founder recites it. And a proven concept still sat in public view for a year and a half while an industry of professional winner-finders scrolled right past.
The word doing all the work in that cliché is execution, and most people hear it as game craft. Learnings' Meowdoku game is two old mechanics and a cat.
The edge was everything around it: a creative machine running thousands of concepts and smart variations with humans + AI, someone actually reading the results and closing the loop of ideation → generation → analysis fast, until more and more winning concepts were found and UA spend scaled massively.
But even with the technology to achieve this machine, it’s clear that some humans are getting much more out of it - and I would call it being thoughtful and data-driven coupled with game tests - actually thinking about the smallest details in what could make for a fun playable game.
That’s the thing at the core of Sett, not only having a machine of AI agents that supplies you with winning creatives at scale, but having the most thoughtful, data-driven, and expertise based AI agents that work for you to create quality and really find the fun (and hence UA scale) for your game playables.
Your next winner probably already exists. On a chart you skimmed this morning, in a competitor's teardown, in the concept your team parked last quarter. Ideas will keep sitting in the open, free to anyone. The machine you point at them is the thing nobody copies overnight.
See you next time.
Yours, Jonathan (still not wrapping anything in cats) Fishman
About the Author
Fishi is the Head of Marketing at Sett. His brain is a chaotic jukebox of ideas with more cultural references than any feed can handle. He collects sneakers and plays chess while you’re still counting sheep.
