In the past, a game marketer’s job was akin to repetitive, manual labor: planning and creating ad assets from scratch, setting them up individually for every media channel, and monitoring performance one by one. However, as generative AI and Agentic AI rapidly automate these processes, the marketer’s role is shifting from 'executor' to 'strategist who determines what to do and designs the direction.' The race to create more assets faster has reached its limit; now, the ability to accurately measure why certain assets work and to make informed decisions is becoming far more critical.

Appier is one of the companies at the forefront of this change. Since its founding in 2012, it has grown as an AI-native Agentic AI as a Service (AaaS) company, with a name that combines 'AI' and 'Happier'—reflecting its vision of a world where everyone is happier with AI.

At the 'Game UA 2026' seminar on June 4, Appier introduced its 'Creative Intelligence Loop,' which focuses on its Agentic AI capabilities. It is a virtuous cycle structure where, once a client inputs their video assets, the AI learns from the campaign data to create new assets, executes them as actual ads, and then autonomously learns from real-time performance to optimize the results.

Following the seminar, INVEN met with two general managers from Appier Korea—Lee Bo-hyeok, GM of Ad Cloud Solution Sales, and Park Seon-gyo, GM of Account Management—to discuss Creative Intelligence and the future of marketers in the AI era.

To start, could you introduce yourselves and share your perspective on the current situation where AI has taken center stage, given that your company has been working with AI since its early days.

Lee Bo-hyeok - I oversee sales at Appier Korea and have been with the company for about eight years. I joined when we were starting our ad business and have been through Appier’s growth, including the launch of our CRM solutions.

Appier is a company that helps clients achieve their KPIs through AI. Since our CEO, Chih-Han Yu, was a researcher in autonomous driving, we have integrated AI into everything—from ads to CRM—from the very beginning. The biggest change is that while AI used to work invisibly in the background, such as in campaign optimization, generative AI now allows us to create ad assets directly and show them to clients. In a sense, the invisible AI has now emerged to help automate client workflows.

Park Seon-gyo - I’ve been with Appier for about two and a half years, having previously worked at an advertising agency. Initially, I led the Korean operations team, handling ad optimization, asset planning, and performance analysis. Since June of last year, I have also been overseeing the U.S. and European regions, sharing best practices with the global team and applying international insights to the domestic market.

Developers I’ve interviewed recently have complained about rising ad costs and creative fatigue. What changes do you see Korean game marketers experiencing, and are there any unique characteristics in Korea compared to the global market?

Park Seon-gyo - The biggest difference is that Korean marketers invest much more time and effort into the creative itself. They delve deeply into the entire process, from planning and design to production and reviewing which assets performed well. In the global market, while assets are important, there is more focus on how to better utilize and maximize media channels, and whether those channels are truly beneficial. One side prioritizes the creative, while the other prioritizes operations based on media characteristics.

Lee Bo-hyeok - Preferred genres also differ by region. Korea particularly favors MMORPGs and idle/raising games, while global markets like the U.S. are strong in match-3 and casual genres. Consequently, differences arise in how Korean and global marketers plan their assets based on these genres and preferences.

You mentioned global insights like MAU Las Vegas at the last seminar. Are there areas in the global landscape where Korean game companies are still lagging behind?

Park Seon-gyo - It’s not so much that they are lagging, but rather that the Korean market is not as large as the U.S., so there are fewer media options available. When you build a media mix in Korea, it narrows down to a few representative channels, whereas overseas, there are far more channels to utilize. The approach is different, and the biggest difference is the focus on 'incrementality'—the actual added effect of a specific media or ad, excluding the performance that would have occurred anyway.

Overseas, they scrutinize whether adding another channel is just reaching the same users again or if it provides additional value. While Korea focuses more on assets or user interest, the global market focuses on incremental effects. Leading domestic companies are already looking into this, and I expect mid-sized studios to follow suit.

Lee Bo-hyeok - Because the population and market are larger overseas, there is a huge variety of media. In contrast, when people talk about performance media in Korea, it is concentrated on a few large channels, and it is rare to see active use of programmatic channels—a sentiment shared by both agencies and game companies.

Does the tendency of Korean gamers influence why Korean marketers focus so much on assets? It seems the frequent appearance of celebrities in mobile game ads is an extension of that.

Park Seon-gyo - Korean gamers love MMORPGs and place great importance on content elements like gacha systems. They care about whether those elements are well-integrated into the assets. Another important factor is the 'sense of trend.' They want to see if a game is truly successful and worth playing with friends. Korean marketers seem to worry a lot about how to creatively express that game content and sense of trend. The effective assets vary by genre: puzzle games provide immersion by letting users experience the gameplay, while blockbusters highlight the storyline or the charm of the characters.

Lee Bo-hyeok - Since Korea values IP and blockbuster titles, hiring celebrities is effective. That is why overseas game companies also make sure to use celebrities when marketing in Korea.

Park Seon-gyo - That can also be linked to the 'sense of trend.' If a developer spends enough on ads to hire a celebrity, gamers perceive it as a signal that the company has high expectations for the game and will service it for a long time. And as the expectation for long-term service grows, so does the confidence to invest (spend money).

In your last lecture, you emphasized the importance of differentiating assets by media using the analogy, 'You wouldn't put Son Heung-min in goal.' Have you seen actual performance differences between using one 'winning' asset across all media versus differentiating by media?

Park Seon-gyo - As we are on the media side, it is difficult to comment on direct performance differences. However, we do notice a difference when clients provide assets: some give us the exact sizes used on other media, while others create assets specifically for Appier. The market is still divided. Some marketers prefer diverse strategies, while others apply insights from one channel directly to others.

Since the market is currently geared toward large media channels, marketers' standards are set accordingly. However, media like Appier operate differently and require assets tailored to that environment to work well. To break that mold, we introduced a solution that automates asset planning and insight generation using AI-generated assets.

Could you explain the 'Creative Intelligence Loop' of the solution you introduced in the lecture in more detail?

Park Seon-gyo - The first step of the solution is for the client to input their video assets into our system. The system then reads the campaign data of those assets to identify which elements worked well and creates primary AI assets based on that. Next, we run actual ads and look at performance metrics like CTR or Cost Per Install (CPI).

Another data axis involves analyzing what elements are in the video—its length, whether it uses characters, which colors are emphasized, and whether the storyline is character-driven or reward-driven. The key is to combine these campaign data axes with video element axes to score what is most effective, and then iteratively produce secondary and tertiary assets based on those results.

This is a process many marketers are already doing, but the key message is that if an agent does it instead, you can achieve similar or better efficiency with less of your own time. This creates a virtuous cycle—the 'Creative Intelligence Loop'—where AI doesn't just stop at creating assets but autonomously learns from real-time performance to optimize them, continuously improving within that loop according to the brand.

At the Las Vegas roundtable, you shared the insight that 'the difference between high-performing teams is not better assets, but better measurement infrastructure.' How is this measurement actually carried out?

Park Seon-gyo - When I was at an agency, one of the things marketers spent the most effort on was naming and labeling assets. For a character introduction video, we would add labels like 'CharIntro_CharName_30s' to extract insights.

While this can be done manually, the VLM (Visual Language Model) in the solution we introduced does that work instead. Now, instead of humans manually extracting elements and cramming them into a file name, they can leave that to the AI and focus more on whether the actual performance is accurate, whether the methodology is correct, and how to improve it.

For measurement to be valid, enough data must be accumulated to form a confidence level. Therefore, the more marketers use the solution, the more the model improves. The method is what is commonly known as A/B testing. The model continuously verifies in the backend whether performance improved specifically because of a certain element, while excluding other variables between element A and B.

Tests that would take humans weeks or months to do one by one are prioritized and conducted by AI, which also provides statistical confidence levels. Above all, this data is not a prediction; it is based on the actual campaign performance of AI-generated assets and the actual video elements extracted by the VLM, making it closer to real-world measurement than speculation. Our message is that when the excellent content capabilities of Korean marketers are combined with this accurate measurement infrastructure, it can create even greater synergy.

It was impressive that you mentioned that even the timing of when the 'X' button or 'Play Game' button appears is a strategy. Could you elaborate on that.

Park Seon-gyo - People have different patterns and psychologies for each platform. The key is when immersion is at its highest, which is usually the moment an action is required. On YouTube, it’s when the skip button appears; on media like ours, it’s when the 'X' or close button appears—immersion rises at the moment the user is prompted to act.

You need to understand when to trigger user action within that small mobile screen. That’s why I mentioned that for the CTA at the end, how long it is exposed is more important than just showing it. A common-sense understanding of how platforms are used allows data-driven approaches to connect naturally.

If AI learns brand assets to automatically generate content, how is quality control, such as brand tone and manner or guideline compliance, handled?

Park Seon-gyo - We ensure that content is generated only after passing through the most basic layer of brand guidelines, brand safety, and tone-and-manner rules provided by the client, so they don't have to worry about the basics. However, since it is AI, mistakes can happen, so we inspect in three stages. First, humans inspect manually; next, humans and AI inspect semi-automatically, selecting only what needs human review; and once this process is sufficiently improved, it is inspected entirely based on AI and systems.

In addition, we have set up a mechanism where all AI-generated assets are shown to the client for their approval or rejection. Even if the AI creates the asset, the final decision-maker is the human—the client and the marketer in charge. Furthermore, the approval and rejection themselves become learning signals. Assets in a direction that is frequently approved are created more, while rejected types are avoided in the future, continuing to learn within the loop.

Are there any creative intelligence elements specific to the game industry, unlike general advertising?

Park Seon-gyo - Game developers often aim for global expansion beyond just a Korean launch. Therefore, how quickly and precisely they localize is important; they must react sensitively not only to text copy but also to the story and content within the video to suit the local market. Because the solution we introduced extracts elements to explain why an asset works and suggests directions for secondary and tertiary production, it is particularly useful for game developers.

Recently, there has been significant fatigue among gamers regarding low-quality AI-generated assets, exaggerated ads unrelated to the actual game, and indiscriminate ad exposure. Does the method of rapidly generating large quantities of assets centered on performance increase this fatigue, and how do you balance short-term performance with assets that accurately represent the game.

Park Seon-gyo - First of all, such assets are used because they actually work in the market. If people didn't react at all and it didn't lead to sales, they wouldn't be used; there is a part that aligns with human psychology to create conversions and is verified by data. However, I agree that fatigue increases if there are too many such assets. That is why our solution includes a mechanism where the marketer decides whether or not to use the asset.

Also, interestingly, if there are too many AI assets, a backlash can occur. Empathy and connection toward what is more original, more creative, and what feels like "a human made this well" can grow. Then, marketers get an opportunity to think about original ideas among the assets the AI produces. Mass production is not necessarily bad; it is ultimately a matter of how you utilize and control it.

Lee Bo-hyeok - If an ad using content unrelated to the game—like so-called 'fake ads'—gets a good response, there are cases where it is developed into actual in-game content. In that sense, marketers are becoming more involved in game planning, so I don't think it should be viewed as purely negative.

You mentioned that if AI reduces repetitive tasks, there is more time to invest in other work. In the AI era, what roles remain that only human marketers can and should perform?

Park Seon-gyo - Personally, as my role expanded from staff to senior and then to lead, I felt my work shifting from execution to making better decisions. As AI agents become more sophisticated, the role of individual marketers will likely change in a similar way.

Basic operations and report writing will be left to AI, and the role of decision-making—such as which campaign to look at today, what performance to report, what suggestions to make, and why to choose one of the three options provided by an agent—will grow. Another role will be a kind of 'tuning'—maintaining and repairing the agent so it runs well and inputting better data and signals.

Lee Bo-hyeok - It is difficult for AI to grasp the vision or company culture of a specific game studio. Deciding what vision and goals a title should pursue is an area for marketers who understand the company culture. Marketers will have a greater role in capturing the overall context and making decisions from a big-picture perspective, and by handing off repetitive tasks to agents, they will have the leeway to focus on that.

AI produces results based on data, but there are clearly parts that cannot be grasped by data, and it is ultimately the human marketer's job to synthesize those and provide direction and guidelines.

What do you think will be the biggest change Agentic AI will bring to game UA in the next 1–2 years, and what should Korean game companies prepare for?

Park Seon-gyo - For the next 1–2 years—actually, starting next month—the role of team leaders is to build an environment where team members can solve problems and experiment with AI. Using agents has become as common as using the internet, and the important thing is whether the agent I created can interact with the agent created by a colleague.

The key will be how to increase the efficiency of the entire team, not just my own, and how the team leader designs the infrastructure to achieve that. Appier also operates multiple agents, and they must collaborate under the same goal rather than working separately to create synergy. If goals conflict, it eventually collapses.

Therefore, both individual and organizational capabilities are becoming important, and the ability to make agents interact under a single purpose will become more important than just creating many agents.

Lee Bo-hyeok - I recently met a game company that said they automate repetitive tasks like soft launches and asset/region settings company-wide while churning out 3–4 titles a month. When a new title is released, assets are automatically set up across various media, allowing them to quickly grasp performance and pivot immediately. Tasks that marketers used to do manually are now left to agents, making things much more efficient.

As an aside, even within Appier, we consider AI utilization so important that it accounts for about half of our KPIs. Our sales organization also uses a sales agent that analyzes client meetings and points out what went well and what needs improvement, and the accuracy is quite high.

Lastly, do you have any words for game marketers.

Lee Bo-hyeok - From a sales perspective, solution companies like ours can cover areas that 'Walled Garden' media like Google and Meta cannot. Whether it's in terms of assets or custom tuning tailored to KPIs, we are open, so I encourage you to test as many diverse media as possible.

Park Seon-gyo - In a similar vein, I hope you don't approach Appier with the same mindset you used to operate large media channels. Appier has a different way of operating, and the points where we can maximize results are different. While it is good to bring over existing learning such as machine learning optimization or bid adjustments, I hope you will also consider the direction suggested by our operations team and build things together.

Every media channel is different, and since this is a solution provided based on insights we have accumulated in the advertising market for over 10 years, I hope we can align and work together accordingly.