Your best playable is dying. Now what?: Webinar Recap and Notes
This is a recap of Sett's webinar with Anton Neznamov (Creative Director, Madbox) and Mélanie Sursin (Senior Creative Producer, Madbox), hosted by Jonathan Fishman (Head of Marketing, Sett).
TL;DR (or watched...): Your best playable dies and iterating it almost never brings it back. Madbox generated 40 variations of one playable overnight and none of them beat the original. The work that matters happens upstream: a grid of player motivations against your own game's pillars, production cheap enough to fill it, and the product team in the room, because a playable is the first thing a player actually plays. AI sits in that process as the editor of ideas, never the author. Read the full recap and watch the recording below.
Every UA team knows the feeling. A playable carries your spend for months, then the numbers start sliding and nothing you do to it helps.
We spent 40 minutes on what happens next with two people who deal with it for a living. Anton Neznamov is Creative Director at Madbox, in charge of every UA asset the studio makes, and he worked on video and playables at Playrix and MY.GAMES before that. Mélanie Sursin is Senior Creative Producer, running creative strategy across the studio's titles with agencies, freelancers and the internal team, and she worked as a motion designer and lead creative strategist for Voodoo and Popcore before that. Right now Madbox focuses on three titles, and Pocket Champs takes up to 80% of the effort.
The conversation kept landing on the same thing. Finding the next concept is a production problem long before it is a creative one.
Where ideas actually come from
Both Anton and Mélanie pushed back on the idea that concepts arrive by inspiration.
Mélanie splits the weekly process in two. Around 30% is benchmark work, looking at what competitors and the current top games are running, including games in other genres. A racing game can take a theme, a character treatment or a color direction from a match-three title without copying anything playable. The rest is the team's own brainstorming, plus sessions with the game developers about mini-games, skins and mechanics already in the build that an ad could borrow. They run it weekly so ideas stay fresh.
Anton's contribution is a way of making it systematic. Put player motivations along one axis, collectors and achievers and customizers and explorers. Put the game's own pillars along the other. For Pocket Champs that means competition, character skins, biomes and the reward gadgets with their rarity. Now you have a grid of cells, and each cell is a concept nobody has built yet.
How deeply you fill any cell comes down to production capacity, which is where the conversation went next.
He also makes room for the unsystematic version. You saw a film, you walked the dog, something landed. The grid exists so that those sparks get caught instead of evaporating in a Slack thread.
Copying the market, and when it stops working
Nobody in the room pretended studios don't look at each other's best ads. The question was when it pays.
Madbox tried it for Pocket Champs, a game without a real lookalike, and it did not work. On a smaller title sitting closer to 4X, they took a popular resource mechanic, the collect-process-sell kind, and that one did work.
Anton's rule is about distance. Run a well-spread performing concept when your game sits far away from it and you will win the click, then lose it on retention and conversion, and ROAS is the number you answer for.
There is a version that survives the gap. The heavy 4X survival games running idle-arcade playables work because the loop underneath is the same loop: kill the enemies, take the resource, feed the people, build the tower. Casualized, simplified, still recognizable. What does not survive the translation is a first screen with forty buttons on it.
A playable is the first thing they play
This was Mélanie's closing point and the one with the most practical consequence.
"Whatever works in the video doesn't necessarily work on playables." Different placement, different attention, different job. Your best video is a reasonable place to start looking for a playable concept and a bad place to stop.
The reason is what a playable is. It is the player's first interaction with the game, a smaller and simpler version of what they get after they install, doing the work of a tutorial before anyone has downloaded anything. That makes the product team a necessary partner on playables in a way they are not on video, where you can afford to be stranger and louder.
How they test
Testing came up as the part every studio does differently. Anton put it plainly: take ten companies and you will get a hundred ways to test videos and playables.
Madbox pairs a new playable with one of their best videos, because for them the combination consistently outperforms either running alone, and sends it into a live campaign on AppLovin. Then they watch how much spend it can absorb and decide from there. "Let the Hunger Games begin."
They also run playables solo for cleaner data, though that read is often too thin to be statistically meaningful.
The honest gap, and it is an industry gap rather than theirs alone, is inside the playable itself. Some ad platforms block external calls, some have tracking built in, and very few teams can yet say which second of a playable lost the player. When a concept underperforms, telling a weak idea apart from a weak implementation is still mostly judgment.
Iterating a dead playable is usually a waste
The sharpest moment of the session was about what not to do.
When a playable dies, the reflex is to iterate. Change the palette, change the biome, move the button. Madbox ran that reflex all the way out on a climbing playable that had been working for them, generating up to 40 variations in a single night: environments, backgrounds, difficulty, skins, the number of clicks before the store. None of them beat the original.
Anton's conclusion is the useful part. If 40 attempts at the details cannot move the number, the details were never the thing, and you can stop being afraid of losing a winner because you picked the wrong color of leaves.
They rarely run variations up front at all. The concept goes into the playable whatever the biome, the lighting, the colors or the props, because a race is a race, and if you climb, you climb. Get the main mechanic clear, get the game design working, put it in the market. Iterate hard once it performs, and not before.
AI in ideation, and where they draw the line
Both sides of the call put AI in the same place in the process, which was not where the audience might have expected.
At Madbox it does not generate the idea. The team writes the idea down, roughly, and AI elaborates it against a knowledge base of previous briefs so that everything going to agencies stays consistent. Anton's line: "it's the editor for our ideas, the technical writer, but never the author."
Production is where they let it run. The team has moved from generating individual assets to building playables from scratch, with 2D and 3D assets coming out of generative model APIs and a headless Blender driven in natural language. The catch he names is that you still have to validate the output twice, once as art and once as experience, because no model can tell you how something feels to play.
From our side, the argument was about where the advantage moves. Producing content keeps getting cheaper and faster. When execution costs little, the studio with the better idea wins, and ideas come from systems rather than from waiting.
Where else to look
The last stretch was the audience's question: where do you find ideas that are not on the top charts.
Anton's answer was to widen the room. Invite the product team into ideation, invite QA, invite people who do not work on ads at all. Ideas are the resource now, so let them flow from everywhere and document every step so nothing is lost.
My own MO here is to sometimes leave software entirely. Tabletop games are full of the best mechanical design on earth, the puzzle shelf especially, and almost none of it has been near an app store.
What to take into 2027
Build the grid of motivations against your game's pillars, so exploration becomes a system, not a guess. Get your cost per concept down, because that number decides how much of the grid you can reach. And when something dies, go find the next concept instead of spending another week on the color of the leaves.
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.
