UA Creative Strategy in the AI Gold Rush: Webinar Recap and Notes
Watch the webinar below, or feel free to read the recap and notes!
Everyone has the same pickaxe now.
Meta's creative tools. Veo. Nano Banana. Midjourney. The AI production layer is available to every studio with a subscription. And the industry responded exactly how you'd expect. Volume exploded. The top 2% of gaming creatives still command 53% of all ad spend. The other 97% still underperform. Win rates didn't move.
We hosted a webinar about this with two people who run UA at scale and don't sugarcoat things. Gonçalo Martins, VP of Marketing, UA & Monetization at BoomBit, manages one of the largest UA budgets in mobile gaming. John Wright, CEO of Turborilla, built the Mad Skills franchise to over 200 million downloads and previously ran a 40-person creative studio at Kwalee.
The conversation went places I didn't fully expect. Here's what came out.
Half of BoomBit's Winners Are Already AI-Generated. They Haven't Cut a Single Person.
Gonçalo dropped a number that stopped me for a second: "Around fifty percent of the winning creatives at BoomBit already are either AI generated, at least on the hook side."
Fifty percent. At a studio testing 15,000 creatives a year across a portfolio of games.
But then the follow-up: "We didn't reduce the team by half. Not even close".
This is the part that most of the "AI will replace creative teams" takes get wrong. BoomBit didn't trade people for AI. They traded slow cycles for fast ones. The same team now produces more, tests more, and iterates faster. AI filled in the gaps that used to cost time, not the gaps that required taste.
John Wright saw the same pattern from the other side. At Kwalee, he managed forty people, "40 people, that were creating new creatives every week. Video editors. Playable engineers." The mission was always the same: find a hero creative. A single ad that captures more than 50% of spend for a game, a country, or a network.
With AI in the mix, "one video editor can now do a hundred times more than they did previously." Production capacity went vertical. But the bottleneck just moved.
"The issue with that," John said, "is the bulk of the AI creation tools are not good enough to go to market without heavy editing."
Good for raw material. Not good enough for the final mile.
The New Bottleneck: You Can Produce Thousands of Creatives. You Can't Test Thousands of Creatives.
This was the thread that ran through the entire conversation. AI solved the production problem. Nobody solved the testing problem.
High-spending gaming apps now produce 2,743 video variations per quarter. Some studios like Kingshot deploy thousands of creatives per day at peak. The supply side is handled. The demand side, meaning network capacity for testing, hasn't caught up.
The networks won't let you push all of them live simultaneously. There's a ceiling on how many creatives you can test per week per account. So BoomBit built a workaround.
"We start using platforms like Facebook, but just go back to the creation of those very small campaigns," Gonçalo explained. "Like five dollars just on Brazil, just to get IPM. Just to get CTR and conversion rates, even ignoring ROI."
Five dollar campaigns in Brazil. Not to drive installs. Just to filter signal from noise before committing real budget.
It's imperfect. Gonçalo acknowledged that roughly 25% of potential winners will slip through a filter this coarse. "You're gonna probably lose them because these indicators are not perfect. But if you don't do that, it's impossible to test a thousand, ten thousand creatives."
The math is simple. If you can produce 10x the creative volume but your testing infrastructure stays the same, you've just built a bigger pile of untested ideas. The studios figuring out cheap, fast signal extraction from that pile are the ones pulling ahead.
UA Managers Are Now Making Ads. That's New.
Something interesting happened at BoomBit that I think signals where this whole role is going.
Gonçalo described a change in his own workflow: "Whenever I have an idea that I know that is AI-able, that a classic Veo or Nano Banana can deliver something, I cannot even nowadays wait to send it to the team, write the script, wait for them to produce, send me back, then it's not really what I had in mind, change it."
He's going directly into the tools. The VP of marketing, making ads himself, because the feedback loop with the creative team is too slow for certain types of ideas.
And it's not just him. "The recent creatives for Big Helmets, they were done by our UA manager, just completely by her." A UA manager. Not a designer. Not a video editor. Making the best-performing creatives for one of BoomBit's games.
This would have been unthinkable two years ago. UA managers ran numbers, managed campaigns, analyzed performance. They didn't make ads. The tools didn't allow it.
Now the tools allow it. And the result is a flatter, faster loop between insight and execution. The person who sees the data and the person who creates the next test can be the same person. For hooks especially, where speed and iteration matter more than production polish, that loop compression is a real advantage.
BoomBit formalized this. Where their creative pipeline used to run "three concepts per game in two weeks," it now includes a mandatory AI hook vertical. "One gameplay, one new idea for a new character, and then like five hooks. And that right away is transforming a little bit how we are producing."
Three concepts became eight. Same team. Faster cycle.
AI Can Make Hooks. It Cannot Make a Game.
There's a critical distinction that gets lost in most AI creative conversations. John Wright broke it down by creative type:
Static images? "No problem." AI handles these well.
Video hooks? Strong. BoomBit and Turborilla are both using AI to generate six-second hooks that get stitched to best-performing gameplay footage. "The AI can produce all the hooks. This we're doing already in our studio," John confirmed.
End-to-end videos? "Still requires a lot of editing." The output needs human finishing before it's market-ready.
Playables? Gonçalo was direct: "On the playable side, it's still very far from AI making anything from end to end or even making anything."
The reason is fundamental. Video generation tools like Veo and Nano Banana can produce convincing visuals because they've been trained on massive visual datasets. They can approximate your game's look.
A playable ad isn't a visual approximation. It's a coded interactive experience with game mechanics, progression, and a difficulty curve.
"Those models never saw your game," Gonçalo said. "They never saw Darts Club. They never saw Entrails. So they don't know really the real assets. How do they move? What's the game?"
John added the IP dimension: "If you have a brand like Monopoly or Transformers, which is very IP heavy and there's legal ramifications if the logo's wrong... you just can't do it, in my opinion."
The entire AI creative conversation in mobile gaming focuses on video and statics. Almost nobody discusses AI for playable production. And playable performance keeps climbing. Playable ad performance scores went from 51 in 2023 to 164 in 2024 to 191 in 2025. Playable ITI conversion rates are 8-16x non-playable formats.
The format that's hardest for AI to automate is also the format that performs best. That's not a coincidence.
The Chinese Studios Are Running 80 Playable Concepts for a Single Merge Game
When we talked about creative strategy for playables specifically, the conversation turned to who's doing this best right now.
I brought up the Chinese studios. Century Games. Microfun. If you pull up Sensor Tower data for merge games like Tasty Travel, Travel Town, or Gossip Harbor, you'll see something striking. These studios are testing sixty, seventy, eighty mostly different playable concepts for a single game.
Not iterations. Concepts. Completely different game experiences packaged as playable ads, each one a mini world designed to test a different hypothesis about what makes players install.
You can reverse-engineer the thought process behind them. How to simplify the game and direct it to a different audience. How to use a specific slice from the actual game without going all the way to misleading. How to build a difficulty curve inside a 30-second experience that makes you want more.
This is creative strategy that requires deep game design knowledge. It's taste. It's understanding how a player thinks, what makes them lean in, what makes them bounce.
John connected this to the broader evolution of playable ads: "People were building mini games as playables and they were creating these non-realistic core loops which have nothing to do with the game. And then they realized, oh, if I add this as part of the game, then maybe I'll get more traction."
The mini game strategy that's dominated UA creative for the last five years started with playable ads. Teams experimented with mechanics in the ad, saw engagement patterns, and fed those patterns back into the actual game.
Gonçalo described BoomBit doing exactly this: "We do use playables for our games to understand what users like to click more, engage more, and characters that they click and choose more." Using the ad as a testing ground for the game itself.
I brought up Dumbledore on this point: you need to choose between what's easy and what's right. The easy path is templates. Shove your game into a pre-built structure and hope it works. The right path is building a fun experience that's actually about your game.
Templates will keep multiplying. But they won't find the next save-the-king concept. They won't find the next arrows concept. Finding those requires the kind of creative exploration that happens when you lower the cost of building a unique experience, not when you fill out a form.
Creatives Are the Last Lever UA Teams Control. Networks Want That Too.
John Wright laid out the history with uncomfortable clarity.
Ten years ago, UA teams controlled everything manually. "You would look at a particular game ID, and you would say the performance of that game was amazing. I will manually increase the CPI in just the US." Granular, manual, high-control.
Then networks started automating bid optimization. Then ROAS campaigns became the trend. Each step handed more control to the network. Sub-source blacklisting went away. Manual CPI adjustments disappeared. The levers that UA teams used to pull are now inside the network's algorithm.
"The last real lever we have in user acquisition is creatives," John said.
And now the networks want that lever too.
AppLovin recently announced they're automating creative production for ecommerce advertisers, building interactive end cards and planning to expand into games. Meta is doing the same with video production automation. They're starting with ecommerce because it's simpler. No gameplay to capture, no game-specific characters, no progression systems. Just products.
But the direction is clear. John connected the dots: "Five, six years ago, AppLovin, IronSource, Mintegral, all of them had creative studios internally." When playables became a movement, networks built in-house teams to create better-performing ads. The incentive was straightforward. Better ads mean more conversions. More conversions mean more revenue for the network.
"They are influencing the front of the funnel as well as all of the other stuff," John explained. "The amount of impressions can stay the same, but the amount of money they make through converting users will go up."
When the network that distributes your ad also produces your ad and optimizes delivery for the ad it made, the studio's creative autonomy narrows to almost nothing.
Gonçalo pushed back on this for gaming specifically. "On the creative side, I don't see where they're gonna take the advantage. Where are they gonna know more about the games and the apps than we do? We medium to large publishers."
His argument: networks see both sides of the ad ecosystem (which gives them an edge in bid optimization) but they don't have deep knowledge of individual games, their characters, mechanics, or player psychology. For small publishers or ecommerce, network-generated creative makes sense. For studios with real game knowledge and established brands? Less so.
For the next two to three years at least, Gonçalo doesn't see network-generated creative replacing what medium-to-large gaming studios do in-house. But both John and Gonçalo agreed the direction is clear.
The question for every studio becomes: how long do you rely on the network to produce and distribute your ads before you realize the one lever you had left belongs to someone else?
LoRAs: Training AI to Know Your Brand
John introduced a concept most UA teams aren't thinking about yet: LoRAs, or Low Rank Adaptations.
"It's essentially a way to train generic AI to make it more on brand for you," he explained. "Like a brand book. Ten pages of information that says these are the colors you want, this is how you want the characters to be received."
Studios are already using LoRAs to fine-tune open-source AI models so the creative output looks like their game, not like a generic AI hallucination. John has seen it done for video, but not yet for playables.
This matters for a reason that goes beyond production quality. It's about brand protection.
Turborilla's Mad Skills franchise has been downloaded over 200 million times. "We do not want to anger our community and our audience," John said. "So we have to be very careful."
John described a new role at Turborilla: "We're having a release manager for creatives now. Whereas we didn't used to have that." Previously, the person who made the creative was the person who shipped it. Now there's a separate QA layer specifically to ensure AI-assisted creative meets brand standards before going live.
AI doesn't just create a production problem. It creates a quality control problem. When one editor produces 100x the output, someone needs to make sure all 100 are on-brand.
The Anti-AI Backlash Is Coming. Possibly Already Here.
John's spiciest prediction: a segment of consumers will actively reject AI-generated creative, and some studios will turn that into a brand strategy.
"There will be an idiot reaction to the amount of AI creatives in the market, and people will, a segment of people, will be highly turned off or offended by this production."
He's already seeing it in adjacent markets. Clair Obscur, an indie game that won game of the year, had the award revoked because the prototype used AI, even though the final game was built without it. On Steam, there's a growing movement of players discovering and boycotting games that use AI in development.
"I can see that as a trend where non-AI creative becomes a movement," John said.
But John was careful not to make this a blanket prediction. "It depends on how good the AI has become. If the creative production is good enough where people can't tell the difference." The backlash is against detectable AI slop, not against AI that's indistinguishable from human work.
Meanwhile, Supercell is still spending millions on live-action shoots with Haaland. Short dramas are selling millions, but people still want Oscar-winning movies. There will always be a place for the crafted, the intentional, the human-directed creative. As Strauss Zelnick, CEO of Take-Two, put it: "Entertainment was always created by a human and a tool. The tool won't become the entertainment experience."
One Year From Now: AI Agents in UA, Bigger Teams, More Work
We asked John and Goncalo to predict what UA looks like in March 2027.
Gonçalo was confident: "I would be surprised if one year from now BoomBit is not gonna have an AI agent working on the UA side." Not for creative production. For operational analysis. Finding trends across geos, identifying which audience segments convert differently over time, flagging retention patterns that humans miss at scale.
He also noted that some UA managers are already testing OpenClaw for managing Facebook ad accounts. Letting an AI agent upload creatives, adjust bids, make campaign changes. "Very scary," he said. "But the fact that we're already starting to see these movements tells me that in one year from now, for sure, we'll have AI agents working within UA."
His prediction on team size: "I really am not thinking about reducing my team at all. I can only imagine that will bring them more work, more things to analyze, more campaigns to be nitty gritty about."
More work, not less. Because when production gets faster and analysis gets deeper, the number of experiments you can run grows. And each experiment still needs a human to decide what to do with the result.
John agreed. Team composition might shift. The workload and focus will change. But the creative team isn't shrinking. AI agents can handle rule-based decisions ("if the ROAS drops below X, reduce the bid by Y"). The non-obvious decisions, the strategic bets, the taste-driven calls about which concepts to explore, those stay human.
What This Means
The AI gold rush in UA creative is real. Production capacity exploded. Costs dropped. The volume game is played.
But the studios pulling ahead aren't the ones who produce the most. They're the ones who built systems to filter, test, and learn from what they produce.
AI on the ad generation front is a production accelerator. It is not a strategy accelerator. The strategy still comes from people who understand games, understand players, and understand what makes someone stop scrolling.
And on playables, the format that consistently outperforms everything else, AI barely scratches the surface. That's where the opportunity lives. Not in generating more of the same, but in exploring what hasn't been tried.
The gold rush metaphor works because in every gold rush, most people dig the same ground with the same tools. The ones who struck gold had better maps.
For creative strategy, the map is creative exploration of new concepts at scale.
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.
