At Tumblix Studios we make games for Roblox and for phones: a supermarket heist, a dinosaur ranch, a superhero city, a zombie story shooter. They look nothing alike, but they keep teaching us the same thing.

Most players decide in the first minute, and most of our fixes end up being fixes to that minute.

The leak was right after the fun part

Shopping Cart Mayhem gave us the clearest numbers. In the live build, out of 600 joins, 87.7% started a first heist and 76.2% made a first steal. Only 24.3% made it to a first escape. Players were doing the exciting part and then losing, and most of them left right there.

We did not add a feature. We made the first heist easier to finish:

Across 220 new players on that build, first escapes went from about 24% to 75.9%, and 51.8% started another heist after escaping.

Rules we now apply to every first minute

Keep it simple. A kid must get it without reading. We prefer making an existing system clearer over adding a new one, and each new feature gets a one-line "what to do" and a pointer to where to go.

No dead time. We assume a new player gives a game about ten seconds, so something has to keep them busy every moment: a goal, a reward or an event. A light story ties these together as short goals, not long quests or cutscenes.

Teach the how, not only the what. On one build, a first-time player knew what the first task was but not how to do it, and the tasks felt like nothing. They were rings and meters on the HUD with no change in the world. Now each mechanic gets a short card with an animated demo and the exact button word, and nothing counts against the player while that card is open. Every press has to do something visible: the character holds the right tool, the thing being made piles up, a miss has a consequence. Sound is a large part of feel, so we add it early rather than ship silent.

Make each moment satisfying. Build-up then release, a clean completion, a number going up, a short sound. Each beat stays under 3 seconds, can be skipped and never blocks play.

Never tell a player to leave. One of our end screens said, in effect, come back tomorrow, right when the next round was about to start. The end screen now says what comes next, and tomorrow's reward sits in a small line.

Let them think, then help on demand

Pet Dealer has catches that need a trick. Showing the solution up front kills the fun, but silence is confusing too. So the game says only that you need a trick to catch, and a tappable hint gives a vague clue first and a clear one second. A catch made without opening the hint scores double.

We also learned the difference between hard to discover and hard to do. A child who knew every step of a trick still failed it and quit. Discovery can be a puzzle. The doing should not need precise standing, tiny timing windows or one slip resetting the chain.

Phones first, art of our own

Most of our players are kids on phones, many of them cheap Android ones. A game that stutters or looks broken on those phones loses everyone, whatever its features. So phone performance is the release gate, above new features. Our rule is to test on low-end Android first, a Redmi Note 11 with 4 GB of memory, then on flagship phones and the iPhone and iPad, and to keep the download under 200 MB.

The same thinking shaped the art. For the Unity game we bought no asset packs. We wrote our own code, built our own models in Blender, used only public-domain textures as raw material, and kept most textures in small shared atlases, 512 px by default. Custom art made it look like ours, and light art kept it running on the phones our players actually have.

One thing to take away

Before you add anything to a game, play its first minute as a new player on the cheapest phone you can find, and write down every second where you are waiting, confused or about to lose. Fix those seconds first.

Back to the map