How Winners Build A Mobile Game UA Creatives & Playables Machine

There's something appealing about throwing spaghetti at a wall and seeing what sticks.

You don't need to think. You don't need to choose which strands are more likely to already be cooked.

You simply grab a bunch, hope your hand doesn't get burned, hurl them toward the wall, and let that wall tell you the answer: "Is it cooked enough?"

I'm talking about mobile game UA creatives, of course. Not spaghetti.

There's been a ton of talk recently about the importance of creative volume, and that the teams winning are outproducing everyone else on creatives.

While true, volume is a crucial part of winning in UA, it's only *part* of a consistent, winning system.

In the case of creatives, not pursuing systematic discovery of your next winner can and will lead to slop. Be it AI slop or human slop, it won't inherently lead to success. Making changes for the sake of changes leads to very poor UA creative performance, especially when it comes to playable ads.

The core of it all is exploring and exploiting.

APAC Is Running Laps

APAC is home to some of the world's greatest UA & Creative teams right now. They are definitely several years ahead in how they build games, run LiveOps, monetize, and acquire users.

In the case of Gossip Harbor, no one can argue with a chart that looks like this.

I won't go into all the factors that drove that scale. This success is built one layer at a time, and you can't really ignore the monetization success or separate it from the rest.

But we can examine and learn how they don't just throw spaghetti at the wall when exploring new concepts.

Visualizing the Machine

Looking at the past 30 days, AppMagic spotted 407 new creatives Gossip Harbor used that included a playable. These include video+playable units that used the same playable but with a different lead-in video.

The vast majority of these creatives, roughly 70-75%, are variations and iterations of their winning concept that you probably know: save the family.

In a nutshell, they explore the creative possibility space systematically.

At some point throughout Gossip Harbor's life, they landed on the winning concept. Probably after massive testing. Once they landed on that area in the creative exploration space, they started iterating and creating better playables that are more fun, make people play longer and engage deeper, and improve conversion rate.

These are smart and deep iterations. Not just small background changes, color changes, or mindlessly changing the item you're merging.

These are iterations that test valid hypotheses and work to make the playable more performant: higher CTR, better engagement, better conversion rate. You'll see this below.

At the same time, they know a concept can't hold forever. In Gossip Harbor's case, once they landed on this concept, the world was flooded with similar creatives from a ton of other games.

At some point, the world will see this concept too frequently and start to become indifferent.

Because of this, the Gossip Harbor team takes the creative possibility space and starts aiming at far and wildly different areas. You'll see this below as well.

When you have a system that learns from data, these stabs aren't random shots in the dark. They're intelligent stabs aiming at areas likely to have correlation with the actual game. In Gossip Harbor, there are all the components you see in these playables: a deep story with a mom and a child (and men), a restaurant and serving customers, classic merge gameplay, and a renovation and building component.

Creativity here, driven by data and knowledge of the components of this game, leads to very interesting concepts they test.

Because of these test hypotheses, a major part of this machine is placing bets. Exploring the immediate area around them. Perhaps the playable wasn't fun enough and failed because of that, and with minor changes they'd actually have a new winner.

More on that in the examples below.

---

WINNING CONCEPT AND ITERATIONS

How does all of this look in practice? Here are a few examples I think you'll find interesting.

The Main Concept

In the main concept, you see the girl and her mom whom you need to save by fixing the broken home they live in. You do that by tapping on different areas to fix, merging items to get what you need, and fixing the place.

Iterating: Making the Playable Longer and Deeper with Three Items vs. One

Exploring the space around this concept, one smart iteration tests whether adding more depth and game content to the playable (letting people complete the entire puzzle and not only one level of it) improves top-of-funnel and bottom-of-funnel performance.

Adding More Story and Choose-Your-Own-Adventure / Episodes Style

Another iteration evaluates whether signaling more of the story/episode element of the playable in the hook section will drive more players to start interacting. This could lead to better performance.

Changing the Theme to Fire vs. Cold

Lastly, in iterations, changing the theme could test whether it drives more users to start interacting with the playable. With a fire theme, they might reach an audience that didn't react to the snow/cold theme.

---

EXPLORING NEW CONCEPTS

At the same time, Gossip Harbor is testing a wide range of new and diverse concepts like these:

Classic Merge Gameplay

Note the valid exploration of just using the core gameplay. In this case merge, and nothing else. Pure merge.

Two things to note here: First, this concept uses a very dynamic and excellent tutorial that guides users when they seem confused. If a user doesn't make a new merge, they get a hint in a few seconds. Also note how the tutorial gradually ends and leaves the user to play. This mechanism was tested broadly across different concepts, including the winning one. This shows how data-driven ideation is probably in place.

Second, this is also a version of the playable testing de facto infinite play, allowing users to play for minutes if they want.

Drag and Merge Combined with Hidden Object

A really cool concept testing whether an audience attracted to hidden object games with some meta merge and serving customers will enjoy Gossip Harbor. Note how many game elements are correlated with the actual game.

They don't just do a random hidden object game and throw slop at the wall. They think deeply, with data, about how this connects to the full game.

Decoration Concept

Trying to appeal to decoration game audiences while still testing whether they enjoy the merge mechanic.

Restaurant Simulation

A zoom-in on a restaurant/cooking simulator that serves hamburgers. Testing whether the audience this attracts will enjoy the restaurant aspect ingrained in Gossip Harbor.

Match-3 Tap with Save the Family

This is the most out-there concept I saw this month. Almost a match-3 mechanic at first glance, which is a very different puzzle than merge. But look closely: this concept, which would probably attract match-3 audiences visually, has a tap-to-match mechanism that is easier. Perhaps they're testing whether that audience, who enjoys more ease will convert to Gossip Harbor.

---

Wrapping Up: What This Winning Machine for Playables Has

**The input:** Data-driven ideas for both exploration and exploitation in an intelligent way, plus deep performance data.

**The output you need to succeed these days:** Volume, diversity, and intelligent iterations. Not spreading your creatives randomly across everything possible, but systematically exploring the space until you hit gold.

Gold that Gossip Harbor definitely hit, and is leading the puzzle category to heights that seemed impossible in 2025.