The Future Is Sett: How to Decide What Creatives to Make Next With AI Agents?
Your life as a UA person somehow became mostly about creatives. Without creatives you're helpless and have no real way to drive scale and performance.
One can blame the networks and the increasing wave of automating almost anything that relates to what UA used to entail, but we're not really about complaining. A much better use of all of our time is to lean into the future and reinvent what we need to succeed in UA for games.
These are weird times to live in. On one hand, AI is advancing at a tremendous pace, and on the other it is not really advancing in the ways we, in the game industry, need it to be. It is a powerful tool, but as a generic tool, out-of-the-box LLMs are not enough to generate value for you, the ones who run UA.
There are many reasons for that, and even though there is progress on some axes, most of gaming is about entertainment. Most of UA is about creating entertaining ads. Entertainment is an unverifiable domain, which makes it extremely hard to train a machine to "magically" generate an entertaining experience for an audience. What is good? What is bad? Who decides?
That being said, there's much agentic AI can help with, and there's much research in the Sett offices that allows us to extract value and shape agentic AI in a way that's useful and valuable for UA teams in studios around the world.
At the end of the day you need to scale UA profitably. That's it.
To do that, you need creatives that perform, both videos and playables, and you need lots of them. The way to find winning creatives, or winners, is by supplying the networks with a high volume of high-quality, diverse new concepts of videos and playables.
But throwing spaghetti at the wall is hardly a strategy, and UA and creatives are full of data and signals that a perfect ideation process would leverage. So this note walks you through how we at Sett are building our own data-driven ideation process with AI agents, and how you can implement it for your studio.
An infinite space, searched by opinion
Let's unpack the problem that ideation is meant to solve.
You as a UA leader need to scale UA on networks that act as black boxes where you can control the input (creatives) and observe the output (performance data).
The hit rate of finding winners is 3.7% for the biggest spenders in the industry, those running over $1M a month, according to NewForm's client data in Singular's recent creative benchmark report. So this tells us one thing: it's practically impossible for humans to predict which creatives will perform on the networks.
In my opinion, there are many reasons for this, and they relate to the unpredictability of entertainment in general. It's extremely tough to understand which experience or messaging will work on a massive audience to become a hit, in film, TV, books, and any other form of entertainment, and it depends on factors we can't really articulate or understand well.
So a UA team like the one you're in, be it on the media buying side, the creative side, or the data and analytics side, is basically trying to search an infinite creative space, all the possible ways to market your game, and hit a winner in it.
The only signal that comes back for you to decide on is the output the networks show you, and translating that data signal into a creative brief is obviously, based on the observed reality, a task the biggest spenders are failing at 96.3% of the time. Crazy thing to think about.
There's just too much data, too much signal for humans to hold and comprehend, and more importantly, too much of a gap between the performance data signal coming out of the networks and the next creative brief, which should be a representation of that learned data in the form of art/creative.
Add ego into this mix and you see the problem. People arguing for hours about which concepts to pursue, where the fun is, and what the data means. That has never produced anything besides demoralizing people and giving most of them a headache.
So instead of promising a secret sauce for "magically" creating winners, we at Sett believe that we can unleash purpose-built agentic AI on this problem, and effectively search the creative space for you in a much faster, more efficient, and better way to find winners, scale UA and get some claps in the next company meeting if that's your thing (or pour an evening glass of whiskey or wine, you do you).
"Data-driven" is a reading problem
One part where agents can help is "reading" the enormous amount of data that is in front of you to make these decisions.
You have your MMP dashboard, your ad managers each with their own analytics, maybe your own creative categorization method, or even synthetic metrics your studio adopted to measure creative efficacy.
On top of all of that you have a brand to take into account, team preferences, and most likely (if you're working with a game that has at least some history) a huge amount of results from experiments that were run, probably documented (at best) in a million different places.
How, on God's green earth, would you expect a human being, as smart as they are, to translate this into a creative brief that is "data-driven"? Especially when the demands on creative directors are to generate an ungodly amount of ideas each week. How can you know that data was actually read, internalized, understood, distilled into actionable insights that are real and statistically significant, and that the specific implementation of these insights in the next brief will actually move the needle?
Well, we mentioned how weird our work is when even the biggest spenders fail to win 96.3% of the time, no shit.
Stop arguing about the next brief. Read the data instead.
See the ideation loop running on your game, with your data.
1) The Sett way: the agent reads the whole market, every day
Sett agents first and foremost understand your game, to the deepest level. What's fun about it: the core loop, the meta loops, the nano loops. What mechanics it has, the difficulty curve, and even LiveOps events.
They also understand your team's preferences, be they matters of taste or brand-related choices.
The agents then understand the affinity between your game and the entire game market, which genres target similar players and similar types of fun, and which adjacent genres are worth looking into.
Once our agents understand the creative space for your game, they read the entire library of ads the market is running, every day, and learn from those that seem to be seen more (signal of winners). Using our own research into making AI understand games, our agents decompose ads into the core blocks that make them: messaging, scene progression, art style, the hooks, and the psychological drivers in action.
So instead of "trying" to spend a huge amount of time scanning the market, each and every day our agents analyze the space as a whole so you never miss an opportunity, be it a new trend or an extremely fast-growing use of a certain mechanic you should adopt.
The idea though is not to just copy competitors without understanding the core building blocks of these videos and playables. The idea is to understand what's transferable to your game and the type of audience that makes high-LTV players for you.
2) Sett agents learn from performance data, generate hypotheses and test them
Now Sett agents also learn from performance data and signals coming in from your networks and MMP to incorporate the feedback you get from these black-box UA machines.
They understand which concepts are trending, which are showing promise, and which ones are not showing any promise at all.
Sett agents also observe, in the case of playables, how players are engaging with your playables and start to collect behavioral observations on top of simple network-derived performance signals.
This continuous log of observations allows our agents to reason and generate hypotheses that can be tested through creatives. For example, testing a specific mechanic within your playables and whether it leads to higher creative performance, or whether it makes more or fewer users drop off. Or whether misleading players beyond a certain point harms retention more than it benefits CPI.
These tests run automatically, through the next batch of creatives designed to test them, leading to a continuous process of improvement in the search for a winner.
It's like an always-on lab with only one purpose: discover your next winner.
3) Sett agents perform structured and grounded ideation
Once they are grounded in data that is as close to the truth as possible, Sett agents run our own proprietary structured ideation process that translates all this data and these hypotheses into actual new concept ideas, ready to be produced by Sett creative generation agents.
It starts with the targeted audience, the type of fun, the psychology of the ad, the mechanics, the level design, and your gameplay progression, all shaped to reflect the core idea and hypothesis.
This process builds a concept up from the tiniest detail into a fully fleshed-out idea based on the data that was observed and understood, and on your specific game knowledge. This gradual and iterative process creates valid new concept ideas to test that aspire to cover the creative space in the most efficient way.
4) Sett agents deliver new concept ideas you approve, throw away, or level up
Then, after these concepts are developed, Sett agents present their thesis for each concept, almost continuously, to a UA and creative expert. That's the human in the loop, who curates the ideas, makes sure they all make sense, and steers the direction with their expert taste.
These concepts then get approved, improved, or rejected by you and your team and automatically passed into Sett generation agents for rapid production, where they are brought to life in your game art, brand, and style, almost indistinguishable from playables and videos made by expert human artists.
You can see some examples here.
The future of UA
We believe this agentic system we're building will shape how UA is done in the next few years, humans working alongside agents, each one doing what it's best at.
Developed and implemented well, this gives each studio and UA team a data-driven machine that is always delivering new concept ideas, one that lets them explore the space and find winners faster. That's one of the only moats left for teams working on great games and trying to make them into hits.
If you're interested in implementing such a machine and leveraging it for your UA and creatives, you can book a call to see it in action and talk details with our team, and join the almost 100 studios that are using it as we speak.
Keep exploring with Sett:
* How the UA manager became a scientist, and why the creative is now the targeting
* The two tracks every creative team runs: exploit the winners, explore for the next.
* What UA work becomes when your team is a set of AI agents you direct.
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 youโre still counting sheep.