THE STATE OF PLAYABLES & AD CREATIVE INDEX

Yo, what's up?

Welcome to Unpredictable Hits - a newsletter for UA, Creative and leadership folks at mobile game studios who want the truth about what's working in mobile game marketing and growth.

I've been reading (and writing) mobile game marketing content for a decade. Most of it sucks for three reasons:

• Data-free opinions from people selling you courses
• Academic nonsense with zero practical value
• Boring corporate-speak from biased vendors

At Sett, we're doing it differently. We'll stay honest to the truth even when it doesn't serve our business, earn your trust, and respect your time. And we'll have fun doing it, because life's too short not to be weird as hell. Let’s try to suck less. 

Each newsletter includes original research and content we created, plus curated takes (with our commentary) on interesting stuff others shared - so you don't have to sift through hours of content.

We won't sell you anything here. Ever. Maybe sandwiches in the future. We make killer sandwiches.

Let's dive in.

Sett Labs #1: ML-Driven Networks want playables, not your intuition

When Deep Blue beat Kasparov, the real story wasn't the win, it was what came after. Engines made moves that looked "wrong" to grandmasters but crushed humans anyway. Turns out 200 years of chess wisdom was just "what makes sense to limited human brains."

Why it matters for UA: AppLovin's Axon and similar ML platforms work the same way. They don't care about your creative instincts. They need ammunition - diverse playables that let them explore the possibility space.

The shift: In the years post ATT, it seems most performant ad networks are moving quickly towards ML driven systems, with very limited control besides creatives. Pretty much no targeting whatsoever. Now creative is the only manual lever left.

What AppLovin’s Axon, Unity’s Vector actually need:

• Diverse playable concepts that test different core game loops, themes, art, aesthetics, and more
• Volume + quality to maximize ML learning signals
• Playables that are actually fun - game and level design matter more than ever

The weird part: These models might discover your idle tower defense playable converts coloring game players into your 4X game. You'd never hypothesize that connection. The ML doesn't explain itself - it just wins.

Bottom line: Your UA team needs to operate like a game lab, not an ad agency. Ship diverse, quality playables at scale and let the machine find the pockets of players you'd never target yourself.

👉 Read the full piece about ML driven UA Creatives.

Sett Labs #2: The playables explosion: what the data actually shows

By the numbers: New creatives spotted monthly that included a playable grew 50x in two years - from 943 in October 2023 to 47,935 in October 2025. Crazy sh*t.

The winner's playbook: Top 20 grossing games now launch a median of 64 playables per month. The top 30 creatives world-wide? 56% are playables vs. just 30% a year ago.

The video-to-playable ratio: Elite UA teams maintain roughly 1 playable per 16 videos. This creates massive combinatorial diversity - giving ML engines like Axon the ammunition to systematically explore and find high-value user pockets.

Case study - Whiteout Survival on AppLovin (Oct 2025):

• 42 new playable ads tracked
• 15 wildly different concepts (not cosmetic variations)
• 27 concept iterations focused on one question: "How do we maximize the fun?"
• One winning concept? 11 variations testing progression depth, pacing, complexity

The variation game has depth: These aren't red coat vs. blue coat swaps. Each iteration tests fundamental game design hypotheses - adding resource types, changing weapon speed to test progression dopamine, expanding core loops.

Why it works with ML: You won't know why a playable wins. The machine found that a tower defense concept for a 4X game performs best when served within a Cake Sort game amongst others. No human would hypothesize that. The ML explores, finds a valuable audience pocket, harvests it, moves on.

Bottom line: You can't compete with the best without world-class playable strategy. The data is screaming it.

👉 Read the full state of playables report here.

Unpredictable Hits Podcast: Zynga’s Zeynep about creatives, AI, and more - AI scales faster - so your core better be good

The paradox: AI lets you produce hundreds of creative variations overnight. But when everyone can scale content, authenticity & uniqueness are the only moat left.

Sett's CEO + Zynga's Creative Director say:

On volume vs. identity:

• Scale your own unique story, not generic concepts
• Allocate a separate X% of the production budget specifically for "core" exploration - new ideas that stay outside your BAU iteration pipeline
• When everyone has access  to the same AI toolset, your unique DNA is what wins

On metrics that lie:

• High IPM means nothing if they drop off immediately - "you just paid to tell a lie"
• Test new core concepts with different methodology than iterations. They'll "miserably fail" on IPM/ROS initially
• Instead, measure in-game behavior: session length, engagement, time spent
• "Stop evaluating moonshots with the same scorecard as daily busywork"

On creative portfolio health:

• Only one hero creative = unhealthy portfolio that limits scale
• When you have a winner, dissect it to the smallest elements (an apple, a specific visual or loop)
• Rebuild new concepts from those successful organs - don't just iterate the same asset to death

On team alignment:

• There should be ONE marketing team, not UA vs. Creative departments
• Artists need performance data - "creatives are solutions to business problems"
• For data-averse artists, use liaisons to translate metrics into actionable insights

On AI replacement fears:

• AI is like graphic designers moving from drawing tables to Adobe - a productivity tool, not a workforce replacement
• "If your strategy is garbage, AI just makes the garbage scale faster"

Bottom line: Produce at AI speed, but anchor everything to your authentic and unique core. Retention is the ultimate metric for creative honesty.

👉 Listen to the full episode here if you missed it!

Meta Andromeda ML “Brain” Continues to Roll Out and Picks Up Fire

There’s a ton of chatter in social media, podcasts, and pretty much everywhere about Andromeda, Meta’s relatively new ML Engine that is powering parts of its advertising platform - specifically using ML to decide which ads to show users. 

What that’s about: It’s not really new - Meta’s engineering team has been talking about it since the end of 2024. This change is moving Meta in the same footsteps as AppLovin and others walked before (and showed it’s massively profitable to do so) - to eliminate human factors such as targeting and manual tinkering with campaigns in favor of using ML to basically: 

• Allow you to go to a network 
• Define your ROAS or other goal 
• Define your budget 
• Provide a sh*t ton of diverse creatives 
• Let the network deliver growth 

Why are people upset? Well, I think that this chatter of “Meta ads are broken” is just because it gets more visibility than things we already know in mobile game UA for a while, because it’s Meta. 

Most of the anger it seems is about the fact that many teams are not well prepared for the shift toward how buying media works with these new UA machines, and are definitely not prepared internally to manage a process that creates the best input these machines need to work. That’s a stream of high-quality and diverse ad creatives - maximizing the network's ability to learn and find your audience. 

👉 Dig deeper → read this piece about shifting UA from audience to asset velocity.

Liftoff's 2025 gaming data: playables are the moat you're ignoring

The dataset: Liftoff analyzed 1.1B installs, 4.7T impressions, 263B clicks over 29 months. For mobile game advertisers specifically, their data reveals a performance gap that's getting worse, not better.

The playable underuse problem: Liftoff's numbers show an absurd divide:

• Top games: playable impression-to-install rates are 8x those of non-playable creatives
• Non-top spenders: playables deliver 16x ITI rates vs. non-playables
• Yet many advertisers still massively underinvest in playables relative to the performance boost they deliver

Metaplay targeting precision: Liftoff's analysis reveals most casual game installs come from hypercasual and puzzle games. Their recommendation: use tools like GameRefinery to identify which specific publishers drive your installs, then create metaplay ads targeting those players' core motivations (e.g., "thinking and solving" for Candy Crush audiences).

Bottom line from Liftoff: If you're not scaling playables, you're leaving 8-16x ITI performance on the table - and that gap is widening between top spenders and everyone else. 

Source: Liftoff

👉 Dig Deeper - Read the full ad creative index report from Liftoff.

Industry Buzz

EA x Stability AI: EA partnered with Stability AI to co-develop generative AI tools for 3D asset creation, starting with physically-based rendering materials and text-to-3D environment visualization.

Square Enix's AI bet: Square Enix partnered with the University of Tokyo to automate 70% of QA and debugging with AI by the end of 2027. 

Razer CEO's prediction: Razer billionaire Min-Liang Tan says AI will "completely disrupt everything" in gaming within two years, launching Game Co-AI (a computer vision tool that watches you play and offers tips) in late 2025.

Krafton goes all-in: PUBG maker Krafton declared itself an "AI-first company" investing $70M in GPU clusters for agentic AI plus $21M annually for employee AI tools - while Palworld studio Pocketpair's new publishing arm explicitly rejects generative AI.

This week’s Game Radar by our friends at Gamigion: 

• ⚡️ SayGames is testing a NEW Hybrid Combo: Screw + thread puzzle fused into ‘Thread Frenzy’
• 🧩 Easybrain (Miniclip) is experimenting with Screw-puzzle formats:
• They’re masters of fast-scaling smart casual. If Easybrain touches a genre, there’s potential. ✈️
• 🇨🇳 ZPLAY doing Zplay things: Nuts theme + sort mechanic mashed into Nuts Sort Jam Puzzle. Pure freestyle development!
• 🧠 Unico Studio is expanding its own empire: Fifth spin-off in its domain: Brain Test 5. Knows its lane, owns its lane.
• 🏄 Popcore is trying a new 3D Tetris-style concept:
• Block Merge 360: Popcore + 3D always deserves a look :)) 
• 🧩 Seen Playvalve‘s Sort Journey: Travel Puzzle? Worth tracking!
• Parrot Games is heating up: After their TD title, they just launched a polished match-3 called Match In Paris.
• Hong Kong-based NovaSonic launched a quirky Idle TD: Combo Mania: Strange mechanics, but interesting to dissect.
• Yooku Games (Team behind Poker Dreams) is back: This time with a coin-dozer/casual casino hybrid called Paws GO! Fast core loop, very East-Asian feel. Good reference title.

👉 Read more about these new games →  here.

See you next time and enjoy the last days before winter really strikes,

Jonathan "The man who saw too many playables" Fishman