Unpredictable Hits #016: Two Soft Launches Already at $100K a Day + 75 Games Cut to 7 Franchises
Hey friend,
Kids are back in school, so there's time to overthink.
You and I have chosen a very strange profession. Every morning we feed creatives into Meta, Google, AppLovin and TikTok, then stare at a dashboard like it owes us an explanation. It never does. We get a number, a green arrow if we're lucky, and the chance to try again.
That's the war, more or less. A small army of UA and creative people trying to get better at reading black boxes built without the need to explain themselves.
The best people I know love the fight. They pay closer attention, ask better questions, and get a little less wrong every month. That's about as close to mastery as this job allows.
Matej Lancaric is one of those people. Last month he ran 360 creatives for Robot Breaker, made roughly 90% of them himself, and turned the whole mess into a five-line production plan. Every line came from something that had already won.
I love that. Three hundred and sixty attempts went in, and five useful decisions came out. Somewhere in that gap is the whole job.
I wrote the first piece in this issue about how AI agents can help us fight this war, because no human can read every ad, result, market signal and half-useful deck before Monday's brainstorm. When the hit rate for accounts spending over $1M a month is 3.7%, we could use a few more brains on our side.
A lot in this one:
- In three weeks, Madbox's Creative Director and I are going live to talk about what you do when your best playable dies.
- We pulled Zeynep Cataner's episode from the archive because her argument about A/B tests killing big ideas is worth your time.
- Matej's full dashboard read, including the two creatives that sent his IPM and ROI in different directions.
- Three games built around AI, plus what it costs when every player action calls a model.
- Two CEOs compare their first hundred days. Nobody launched anything, and one of them cut 75 games down to seven franchises.
- A game dev veteran wrote a genuinely useful handbook for managing AI agents, including everything they are banned from doing.
- Newzoo's 2026 numbers, with mobile growing fastest while downloads keep falling.
- 15 new soft launches, with two already making $100K a day.
Yalla.
Sett Stuff and Analyses
How to Decide What Creatives to Make Next With AI Agents
I wrote this one, and it starts from one number. The hit rate for accounts spending over $1M a month is 3.7%, per NewForm's data in Singular's latest creative benchmark, so everyone with a big budget is failing 96.3% of the time.
Your networks are black boxes. You control the input, creatives, and you observe the output, performance data. In between sits an infinite creative space, every possible way to market your game, and most teams search it by opinion. People argue for hours about where the fun is, nobody wins, and the 3.7% is the score.
The argument I wanted to make is that "data-driven" is a reading problem.
The MMP dashboard, the ad manager analytics, your own creative taxonomy, a brand, team taste, years of experiment results documented in a million places, and a creative director asked for an ungodly number of ideas a week. Nobody reads all of it before a brief. So the brief runs on vibes, whatever the deck says.
The rest of the piece is how we built agents at Sett to do that reading every day, on the market and on your own performance data, and hand a human the thesis for each new concept to approve, throw away, or level up. Almost 100 studios run on it. If you want to see it on your own game, the link is at the end.
Three Weeks Out: Your Best Playable Is Dying. Now What?
Matej's item below is built around the question this webinar exists for: how much of next month goes to squeezing the proven winner, and how much goes to hunting the next one. On September 23 Anton Neznamov, Creative Director at Madbox, and I take that question live.
He brings the studio side. The dying winner is his, and so is the call on what replaces it. I bring the production side, where I watch playables die across dozens of studios.
If you already saved a seat, ignore this. If not: September 23, 2 PM UTC / 4 PM CEST / 10 AM ET, free, no slides, your questions at the end, and the recording lands with everyone who registers whether or not they make the hour.
Worth Going Back To: Zynga's Zeynep Cataner on Testing Big Bets
We're re-running this episode this week, because it answers the question this issue keeps circling: how you test a genuinely new concept without the standard playbook killing it too early.
Amit sat down with Zeynep Cataner, Senior Marketing Creative Director at Zynga, who came up through branding work for Apple, Coca-Cola and Tommy Hilfiger before she moved into games. She brought the storytelling with her.
The idea worth the 49 minutes: she draws a hard line between business-as-usual iterations of a proven creative and core concept tests, and runs them under different rules. Standard iterations get short tests, fast calls and strict IPM thresholds.
Core concepts get more volume and longer learning windows, because a genuinely new idea often brings a different audience whose value shows up in retention and monetization long after the IPM read. Her warning: "If you treat every concept like a regular UA creative, you'll kill your best ideas too early."
The other one: Zeynep calls turning performance data into creative direction "translation." "A creative is a solution to a problem. If the artist doesn't know what problem they're solving, how can they aim?"
Next week: Uri Pearl, VP Marketing at Papaya, is on the show. Stay tuned.
๐ Listen to the episode with Zeynep
Industry Stuff you should know to be better
1. Matej Lancaric Read His Own Dashboard for 30 Days
Author: Matej Lancaric, Brutally Honest
Matej spent a month inside Singular's Creative IQ dashboard for Robot Breaker, the game he runs himself. The month covered about 360 creatives across four networks, โฌ41K of spend and 6,342 installs at a blended IPM of 3.80.
The piece is what happens when someone with taste treats a dashboard as a production planner instead of a report card. Most people never get past the report card.
The mismatch: by creative type, playables sat at a 21.75 IPM against 2.75 for video, so the lazy read is make more playables. Then he flipped to the ROI columns.
On Facebook his character-led creatives beat his plane-led ones on IPM, 1.95 to 1.79, while the plane creatives paid back better, 46% against 36% on cohort D1 return on spend. His summary: "IPM tells you what the algorithm loves. ROI tells you what your business loves."
The โฌ224 concept: his plane concept on Google ran on โฌ224 of spend at a 250%-plus ROI. The most profitable thing in the account was too small to matter. First move there is budget. Then variations to hold it.
The plan: five directions for next month, all tied to winners, including a hybrid where his two best elements finally share one creative. "The dashboard's job is to shrink the brainstorm."
His three reading rules: compare within a network and never across, respect sample size (his best-looking element ran a 35 IPM on ten installs), and watch the actual top three creatives behind every pattern before you brief anything.
2. Three Games With AI in the Core Loop, and the Inference Bill
Author: Max Abrahamsen, Consultant at Naavik
Most of the AI-in-games conversation is about making assets cheaper. Abrahamsen looked at three games that put a model inside the mechanic itself. The pattern that came back is about money as much as design.
MIMESIS (ReLU Games, a KRAFTON studio) is four-player co-op horror where the monster mimics your teammates and replays their recorded voice lines when the game state fits. It peaked at 562K monthly actives in December 2025, its launch quarter, and passed 2M cumulative sales by July 2026 without recapturing that level.
Death by AI (Little Umbrella) drops you into a deadly scenario and gives you 60 seconds to type how you survive. An AI game master decides. Its OpenAI and ElevenLabs stack became financially unsustainable within three days of the Discord launch. Within three months it had passed 20M players.
At peak, Discord says, it was generating roughly 1.2B text tokens and 35M voice characters a day. The piece's phrase for what that means: inference cost becomes part of COGS.
Status (WishRoll, $17M raised) is a simulated social network where everyone replying to you is an AI character with memory. In beta it reportedly cost $12 to $15 per user per day.
Working with Inworld, the team says it cut AI costs by over 95%, then layered on energy, subscriptions and rewarded ads. ARPDAU went up. So did the monetization complaints in store reviews.
3. Two CEOs, One Hundred Days, Zero Launches
Author: Aylin Yazฤฑcฤฑ, Deconstructor of Fun, from Mishka Katkoff's conversation with Jon Bellamy and Alexis Bonte
In 2024 and 2025 two CEOs inherited big companies: Alexis Bonte at Stillfront in October, Jon Bellamy at Jagex the following March. Katkoff got them in a room. The subject was their first hundred days.
The plan predates day one: both arrived knowing exactly what they'd do. Bonte: "The plan was very clear for me from the get-go." Spend the window gathering data and you've burned a third of it.
Subtract: Bellamy divested two studios and killed everything that wasn't RuneScape, then renamed the company "Jagex, the RuneScape company." Bonte cut 75 games across 21 studios down to seven franchises and deleted the regional business units.
Neither launched anything. Jagex then grew four to five times faster than in any year of the previous decade.
One deep cut: Bonte rebuilt his exec team from twelve to six inside a year. People told him it was fast. He says he should have done it in three months.
His line on culture: "We're not a family. We're a pro sports team. If the striker isn't scoring goals, you put him on the bench." Then the kicker: "If you're not decisive, you shouldn't be the CEO." Worth holding alongside that: Stillfront opened the search for his successor on 30 June, and Bonte leads until they name one.
The piece's own claim is that none of this is a CEO problem. The shape of the first months is the same whether you inherit 1,200 people or a team of five, and it's the only window in which the org expects you to change things and forgives you for it.
๐ Read on Deconstructor of Fun
4. The Operating Contract for a Team of AI Agents
Author: Antti Kananen, Gamigion
Kananen started a game project by skipping the engine decision and designing the relationship between himself and the agents that would build it. His first deliverable was a handbook. The part worth stealing is the limits.
Nothing enters quietly: a new engine, paid service, dependency, license or backend doesn't enter the project because it was convenient for one task. It gets proposed, researched and approved. Project state, decisions and the next exact action live in the repo, so a stalled agent or a model switch shouldn't mean rebuilding from memory alone.
The brief: "'make the first tank' is a weak task." The strong version specifies the reference, silhouette and proportions, triangle budgets, LODs, material rules, export formats, allowed tools, forbidden actions and who signs off.
Anyone who has briefed a creative team will recognize the shape.
The rule I'd frame: "Demand evidence, not completion language. A playable capture, repeatable test, manifest and measured comparison are more useful than 'implemented successfully.'"
He's honest that agent-made, game-ready 3D is still young, and that one attractive render can hide a model that fails from the side. The last rule keeps the rest sane. When the process starts optimizing itself instead of the experience, "just play the damn build."
5. Newzoo: $213.9 Billion, and Mobile Grows While Downloads Fall
Source: Newzoo, key numbers from the free Global Games Market Report 2026 (the full free edition lands September 10)
Newzoo forecasts the global games market at $213.9 billion for 2026, up 6.1%, with growth on every platform and in every major region. Mobile stays the biggest at $121.1 billion and grows fastest at 6.8%. Console sits at $46.9 billion and PC at $45.9 billion.
Mobile grows while downloads fall. That's the line for UA people. Newzoo pins the growth on deeper spending per player, Chinese publishers gaining share abroad, and direct-to-consumer channels.
3.7 billion players, and slowing: players reach 3.7 billion this year, up 4.4%, and growth eases to a 3.2% CAGR through 2029. Penetration only nudges from 61.5% to 62.0% of the online population.
Newzoo's read is that future growth depends on retaining and monetizing the audience you already have. China and the US supply 52% of all consumer spend on games.
6. Soft-Launch Sonar: 15 New Games, Two Already at $100K a Day
Author: Jakub Remiar, Gamigion
Remiar's soft-launch sweep is back with 15 new titles, and two stories from recent issues continue. Century Games holds its genre focus with three new merge games and two new 4X games, Lord Rush and Top Force, both already at $100K a day.
RiverGame is back on the 4X map with Top Lords, a follow-up to Kingshot, at $70K a day. In Remiar's words, "4X war is back on the menu."
Smash Fest clones, still coming. The tail we covered last issue is still growing. Supercent joined with its own follow-ups while pushing its puzzle RPG line with Match Clash. Spyke's Combo Games account shipped Cube Smash, a Cubeland follow-up, plus Bubblegum Words.
Fake ads as games: Good Job Games is testing "fake ads" concepts like Police Escape and Spook Escape, which Remiar thinks are being validated for use in the main game. Also in the batch: InnoGames' Horizon Dominion, Miniclip's Rival Nations and SayGames' O! Hole!
Wrap Up
Most teams have enough ideas. Deciding which ones deserve another shot is the bastard. A dashboard can show you 360 creatives and still leave you completely alone with that call, especially when the ad with the best IPM makes less money.
That is the part we are trying to solve at Sett. Let the agents do the endless reading across your performance data and the market, then put a better brief in front of the human making the call. A giant dashboard and a blank page solve nothing.
See you next time.
Yours, Jonathan 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 youโre still counting sheep.
