🎮Using narrative/storytelling with AI to drive 40%+ ROAS lift: with Karen Levy Atias, Director of Creative at Supersonic from Unity✨

Our guest today is Karen Levy Atias, the Head of Creative at Supersonic. After years of building and scaling creative strategy for some of the world’s biggest mobile game hits, Karen has led Supersonic through one of the most dramatic industry shifts: the move from hyper-casual to hybrid-casual. She’s not just theorizing about AI’s role in creative work, she’s using it to produce hundreds of high-performing ads every month without expanding her team.

In this episode, Karen breaks down how the hybrid model is reshaping creative strategy, why chasing the lowest CPI no longer works, and what it means to optimize for ROAS and retention in today’s market. We talk about the power of narrative-driven ads (including one that captured 20% of spend and delivered huge ROAS gains), how Supersonic scaled creative output by 5–10x through AI efficiency, why “human taste” remains the ultimate moat, and how to adapt your testing cadence across SDK and social channels.

Full Transcript Below
Shamanth: I’m very excited to welcome Karen Levy Atias to Intelligent Artifice. Karen, welcome to the show.

Karen: Hi Shamanth. Thank you very much. I’m very happy to be here.

Shamanth: I’m excited to have you because I know you’ve had a whirlwind couple of weeks traveling the world for work. I’m glad we were able to make the scheduling work. I’m also excited because you work on creative strategy at a massive scale for hit games, and there’s so much that you do that I’m excited to learn from and share with people.

Let’s start by talking about the shift in your work over the last couple of years. Your team shifted from hyper-casual to hybrid. What were some of the biggest creative shifts that you noticed as your team made that transition?

Karen: I think the changes and shifts we’re talking about have been happening for a while. It’s not just that hybrid games require this change inside your creative production; the entire industry is changing. It’s a combination.

Of course, we see the industry moving toward hybrid games—games that are generating more revenue from IAP (In-App Purchases). Having a 50/50 model of IAA (In-App Advertising) and IAP forces us to change because we need different users. We need to speak to and drive different users into those types of games. These aren’t just users who play for a certain amount of time, watch X amount of ads, and then generate 100% ROAS (Return on Ad Spend) after 30 days. That is not the case anymore. We’re seeing lower curves where the return takes a while, but the users stay for longer.

We’re seeing this shift in hyper-casual games that are trying to become hybrid. If you had a big hit in the hyper-casual industry in the past—and we know a few that are still in the top charts—we see their monetization strategy shifting, which means they have to shift their ad strategy as well.

The way this impacts creative teams is in how we strategize our creatives and how we approach what to build. In the past, we were looking to drop CPI (Cost Per Install). That meant we could go with a leading creative and iterate on it until we felt creative fatigue because we knew that type of creative lowered the CPI.

Today, we need to level up the sophistication and provide different users with different stories, which means different concepts. We have to think about the story behind the creative and what motivation or emotion I want the user to feel so they will be more likely to click and install the game. Of course, a “good user” is one that will stay and pay for things inside the game.

Teams really need to think about and build this strategy. You have to determine the best approach: Is it a clear strategy built across all your games? Is it something you adapt per game? Do you want your strategy to look at each channel differently? It depends on what you are capable of producing. Can you produce for each channel?

It involves a lot of exploration. In a world where you have endless opportunities and endless things to think about in terms of strategy, it can be very confusing. You need to be very structured and organized. I’m not saying there’s one right way, but teams need to start thinking in a more organized manner rather than just throwing tons of iterations to drop the CPI quickly. Processes are becoming slower. You have to ask: How are you adapting to it? How are you tracking it? How are you learning from it and bringing those learnings back into the creative production team to execute better?

Shamanth: If I understand correctly, that’s been a very big shift. With hyper-casual, you have a bouncy ball game, and there really can’t be too much narrative added to a bouncy ball game, especially when you’re just looking for CPI. But when shifting to hybrid, there is much more depth. You obviously have to look at different channels, and you are optimizing for longer-term retention rather than the immediate click. That seems to be the biggest shift, right?

I know you briefly touched on storytelling and how that drives performance. Can you share any examples of stories or narratives that led to a massive improvement in performance?

Karen: In our creative team, we try to track as much as we can, though it’s not easy to track a lot of creatives across different games and clear out the data. We try to differentiate between the creatives where we are using storytelling—where there’s a thought process and a meta inside—versus the other creatives, which aren’t completely dead. I want to emphasize that: gameplay is still there, and you can continue to iterate and test gameplay. It is important for the process.

However, we’re trying to compare the impact of investing in storytelling—the visuals, assets, thinking, and process—versus the fast iterations where you add a timer or text on top. We have seen in our activity at Supersonic that creatives with storytelling have shown, in most cases, a ROAS uplift of between 20% to 40%. In some cases, we see a 10% to 20% uplift in retention and IAP share as well.

The IAP share is the most interesting part because that is the end of the funnel. You’re bringing in users who are eventually paying, so we’re impacting the game from the creative at the very start. If we see that storytelling creatives are bringing a 10% to 25% uplift in IAP share, that is a big thing.

To give a specific case, we published a tycoon game called Trash Tycoon (or similar) during the year. We had a lot of gameplay scenes inside the game. One of the directions the team decided to take was a more emotional story. You need to clean the city from trash. Cleaning the city can go in many directions—a factory, someone walking in New York City, etc.

They decided to take it in a different direction from an animal’s perspective: how trash impacts animals in the world and how disturbing it is for them. There was a creative featuring a very sad fish swimming in a big sea with a lot of trash around. It really made people feel connected to that creative. That creative immediately took 20% of the share of spend for the whole company. That is very big when one creative takes 20% of the spend across all geos for a title. It was proof that users want to feel related to the story; those emotions make them feel more connected to the game.

Shamanth: That’s so interesting. You would think trash is kind of a funny thing, almost hyper-casual like a bouncy ball, but clearly, people resonate with a deeper narrative about animals more than just cleaning the city. I can imagine that’s a big differentiator because if everyone is making stupid creatives with fake gameplay or clickbait, and you have something deeper, I can see why users would resonate with that.

To switch gears a bit, let’s talk about AI in the workflow. Can you walk us through what your end-to-end creative workflow looks like? Where does AI fit in, what AI tools do you use, and where do humans still add value? I would imagine for the trash example you gave, it’s possible for AI to come up with that, but humans likely need to give a lot of guidance.

Karen: AI is in our processes from the first moment of ideating. I think most listeners use Gemini and ChatGPT for simple things day-to-day. It helps us think faster and shorten our time. So from the moment we start thinking about strategy, AI comes in for ideation and writing the brief itself. It helps shorten the time of writing emails, letters, and briefs.

It also helps us do market research. There’s a lot of data out there, and sometimes it’s hard to come to a conclusion. It is very easy to use AI to give you two or three key outputs from a report you downloaded.

In production, it helps create assets that in the past would require an artist or a technical artist. Creating assets that used to take a few days or specific skills can now be done easily. For example, creating that cinematic theme of the sad fish swimming in a sea full of garbage. Two years ago, I would have laughed and said, “No way, it’s not worth it for me to hold an artist to do motion stuff for this game.” Today, we are doing it with AI.

We use it for creating voiceovers, UGC, and short hooks to pull the user’s attention and stop the scroll. The “thumb-stopping” part is the most important for an advertiser.

However, humans are there for judgment and direction. The AI will not judge the things it is doing. It won’t check where this ad is going to run, if it can be offensive, or if it meets the rules of a specific channel. The AI might take it in a direction of fake gameplay, which I might not want as a strategist. I need to stop that direction and direct it better.

Also, AI doesn’t have taste. It creates something, but sometimes the colors aren’t good or aren’t related to the game. Humans are there to judge all these crazy things. It is time-consuming; sometimes it’s easy, but sometimes it’s hard to bring the AI back on track. Learning to do that is also a skill.

Shamanth: You’re right. It’s easy to overestimate how much AI can do, but being the human in the loop is a very important skill. You need that judgment and taste to make sure AI gives good output. You can’t really let AI go crazy.

Once an ad is made, it goes into testing, and testing has its own challenges. Can you talk about what some of the challenges are with testing that you’re noticing and what your team’s process is for deciding when a creative has proven itself?

Karen: Testing is something where we need to take a breath. It is still in the process of evolving. Creative success today isn’t just CPI. When it was just CPI, it was very easy. Everyone in the industry took a bunch of creatives, tested them on Meta, looked for the CPI drop, and applied the winner to the rest of the channels. Usually, the top creative that dropped CPI would catch on in other channels too.

That’s not the case anymore. Since we’re looking for more depth, the test becomes more complex. To understand the depth of your creative, you need more budget, more time, and more volume. You don’t want to test for 60 days—it doesn’t make sense—but you still want to produce because social campaigns breathe with new creatives. You can’t let them fade for two or three weeks without adding anything new.

You have to find a balance. How much am I willing to spend on new creatives when I don’t know how they will impact overall performance? They might lower my overall ROAS and harm my UA strategy. On the other hand, relying mostly on CPI is just hoping for the best.

We have a lot of channels, and over the last two years, everyone is trying to improve their algorithms. They say you can throw 20 or 30 creatives at it, and the algorithm will do the work. I’m going with this flow because, in a world with AI and tons of creatives, someone needs to decide quickly for us.

However, you need to be structured in your strategy so you can track it. You look for depth, but you can’t waste money or time. Choose your top two or three channels, focus on them, and then apply the learnings to the rest. Don’t use one channel to apply to all others like we did in the past, because there is no “one truth,” and you will miss opportunities. You might miss a creative that isn’t a top performer but drives really good ROAS and specific users. You want this creative in your campaign even if it doesn’t scale massively. You need to spread your risks and strategize according to your game.

Shamanth: I agree. There’s complexity you have to deal with. With CPI, it was just “lowest CPI wins.” Now, maybe the CPI is higher, but the ROAS is higher, or something wins on one channel but not another.

Speaking of channels, I know you guys are on SDK networks and social channels, and obviously, they behave very differently when testing creatives. How does that impact the production and testing cadence?

Karen: On SDK networks, the playable is the platform. The difference between SDK and social is that on SDK, if you don’t have a playable, you’re missing a big incrementality in your UA strategy. It is obvious that you need to run a mix of playables and videos.

Deeper than that, SDK networks are slower in reaction to a new creative than social. On social, I expect to understand the direction of a creative in a week, max two. It’s not common to see a creative on social pick up after one or two months. On SDK networks, you can see those cases. That can be tricky when optimizing.

You need to remember SDKs are slower. When do I upload new creatives? It depends on how much volume you have in your campaigns and how much budget you are willing to put into a campaign where you’re adding new creatives. Are you willing to add a new creative to a main campaign that is spending without testing it before? It’s a question of risk.

Some SDK channels allow you to be confident uploading new creatives knowing the algorithm will kill those that don’t drive the KPIs you optimize for, like ROAS Day 7 or Day 30. You need to learn those channels to trust them.

Compared to social, creatives are the heartbeat. If you’re not inserting “new blood” into the campaigns, you will see fatigue much quicker on social. You’ll see top creatives start to sink. The impact of new creatives on social is very big. Even if it’s an iteration—taking a top creative and doing a different voiceover or text—it is really needed. On SDK, I’m not sure uploading iterations every week will give you any impact.

Shamanth: I agree, SDK is less sensitive to volume. Your team produces hundreds of ads a month. How do you decide how many to produce, how many new concepts versus versions, and how many go into each channel? Is there a rule of thumb?

Karen: In the past, the rule of thumb for hyper-casual was simple: produce X amount of videos and iterations every week, find the winner, drop CPI, and move to main campaigns.

Now, looking at depth, we take it case by case. Each game has its own strategy. The creative producer works together with the growth manager of each title to understand the current need. Are my top channels social or not? If I have a top channel that is an SDK, I need to add a new interactive or playable. If it’s social, I know I need a few new creatives every week.

They build the strategy together every week for every game. They plan how many creatives to build. Are we iterating this week? If we found a new “hero” last week, we don’t need to look for a new concept this week. I will take that new hero and iterate on it because fatigue hasn’t set in yet. I can squeeze the lemon by tweaking the direction while keeping the core concept. The producer takes the UA need from the growth manager and implements it into the strategy for the next week or two.

Shamanth: So it’s very iterative and you’re building on previous weeks’ learnings. I’m curious about the team structure. I would imagine creative producers aren’t always plugged into performance metrics. What does the feedback cycle look like between the growth marketers and the creative producers?

Karen: The biggest interaction these two functions have at Supersonic is a constant update on the status of the game and the UA needs. For example, if we need to push for mass installs and volume, we might aim to buy lower CPIs. The creative producer needs to take this goal and say, “Okay, I need to drop CPI. Let me go back to my experience in hyper-casual and create something scalable.”

This interaction must be harmonious and supportive. If we build a strategy with two new concepts we believe in—based on previous data or market research—the growth manager needs to trust the creative team. They need to feel confident uploading that creative not just to a test campaign on Meta, but to the second biggest channel to see if it drives ROAS, retention, or IAP users.

It requires trust and communication. If the UA manager feels it’s going in a bad direction, they need to talk so the creative team can iterate. If there is no communication, the producer will do whatever they think is good, and the growth manager might refuse to test new concepts. That unbalances the goal of building creatives that drive performance and incrementality.

Shamanth: From what you’re saying, it’s critical that the growth marketer understands what’s being tested and why, so they buy in, and the producer needs to understand campaign performance.

Switching gears again—your producers work with multiple AI tools. How does your team decide which tools to use? Is it left to the producer?

Karen: The world of creative AI can be confusing because a new tool or version comes out every week—VO, Nano, Banana, etc. You don’t know where to focus. In management, we try to keep it focused. We know the things we need: good directions, voiceovers, assets, and short video productions for hooks.

We listen to the people in the field producing the ads. We try to explore different tools and have diversity—we don’t need duplicate tools doing the same thing—but we have specific tools for each need. The producer decides the creative direction and has this “box of tools” to choose from.

They don’t really have limits. They can go in any direction, like the cinematic sea creative I mentioned. We want to buy them the tools to do that. As creative strategists, they need to optimize their work and understand when to use a specific tool. I probably won’t do a cinematic creative every week, nor use the same voiceover every time. They need to diversify. But if data proves a specific tool or voice always works, we go with the data. We allow the freedom to choose the tool and strategy.

Shamanth: That makes sense. It comes down to data, but also the storytelling and strategy that drives the data.

You said you still need a human for taste and judgment. When I talk to people about things AI can’t do, they mention gameplay assets. Are there ways your team works with AI for game assets to short-circuit production, or is that still manual?

Karen: Things are improving rapidly. I try to be careful talking about AI because things change in two weeks. There are tools starting to mimic the environment of the game. If I have a monster in my game and want an animal that fits the environment or lookalike of the game, we are seeing things in that direction.

But judgment is still key for gameplay. AI cannot always control physics—what a character can or cannot do. Does the throw look reasonable? Did the AI mishmash something and make it look weird? Recording a simple gameplay is very easy today, so AI isn’t necessarily replacing that yet. If AI doesn’t make it easier or quicker, what’s the point? I assume there will be a point where you just throw in your game link and it exports level 200 for you.

Shamanth: Meta claims that’s where the world is going—just give us a link, and we’ll extract assets and make ads. That is exciting or intimidating, depending on how you look at it.

Karen: I think judgment will not be eliminated ever—or I hope so.

Shamanth: I feel optimistic because Meta has talked about automation for years—Advantage Plus, etc.—and it’s still not perfect. Using automation completely is risky. I don’t know if I’m comfortable giving the keys to the car to the AI yet.

How has AI impacted resourcing on your team? You mentioned AI helped maintain output without growing headcount. How does this change the skills of creative teams?

Karen: If AI is not efficient, there is no point in using it. Efficiency is the first thing AI creates. To create that fish in the sea, two or three years ago I needed an artist working on animations. Today, I don’t need that recruit. That is the efficiency we are talking about. I am doing more without growing my headcount or skills in that specific area.

Maintaining scale while deepening creative strategy without growing headcount says a lot. We are reaching more and creating deeper work. People are investing time in stories and storyboards, which is more work than just recording gameplay, yet we haven’t grown in headcount. I believe AI will continue to allow creative teams to do more with less, though I hope judgment remains a key factor.

Shamanth: That mirrors my experience. We haven’t reduced our creative team, but the output per designer has gone up dramatically—5x to 10x—along with the quality.

This has been incredibly in-depth. Before we wrap up, could you tell folks how they can find out more about you and what you do?

Karen: I lead the creative growth at Supersonic, managing creative teams and strategy connected to our growth strategies, along with leading social UA activity. I try to lead those two teams in parallel because they have a very strong connection. You can find my activities and what I’m speaking about on LinkedIn. Ping me, talk to me.

Shamanth: Excellent. We’ll link to your details in the show notes. Thank you so much, Karen.

Karen: Thank you very much. It was a pleasure

WANT TO SCALE PROFITABLY IN A GENERATIVE AI WORLD ?