Appier, AB180, Playio, and TikTok co-hosted the 'Game UA 2026: Growth Strategy Seminar Powered by AI' on the 4th. During the seminar, which targeted game and app marketers, Park Sun-kyo, Head of Account Management at Appier Korea, delivered a presentation titled, 'One Creative Cannot Satisfy Everyone: How to Drive Performance with Creative Intelligence in the Age of Agentic AI.'

Park holds a degree in International Business and Economics from Whitworth University in the U.S. and has built a career in digital marketing and ad operations at companies including Echo Marketing, AdCube, and Mobidays. He currently leads Account Management for the KR, US, and EMEA regions at Appier and oversees client management for Ad Cloud solutions.

Park began his presentation by focusing on the upcoming World Cup season. He displayed a photo of Son Heung-min playing through his injury in 2022, noting that experts always highlight the best players in every tournament, with Son being a prime example.

He then posed the question: "If we placed Son Heung-min in every position to get better results, would the team actually perform better?" Using the goalkeeper position as an example, he emphasized that it is a specialized role requiring the use of hands, aerial ball handling, and specific physical attributes like height and arm length. According to Park, while Son is approximately 183cm tall, the average height of goalkeepers in this World Cup is around 189–190cm. This illustrates that even the most outstanding player is not a perfect fit for every position.

Park pointed out that this dilemma mirrors the decision-making process of marketers. Every day, they must choose between a 'one-size-fits-all' approach—where a winning asset is quickly learned and deployed across all media—and a strategy that segments and differentiates assets based on the unique characteristics of each medium.

Park compared the points of peak engagement for paid media users across different platforms. On social apps (Reels, short-form content), concentration is highest in the first 0–3 seconds, whereas on YouTube, due to the skippable nature of ads, engagement peaks at the 5–6 second mark, just as users are about to skip.

In contrast, the programmatic DSP space that Appier handles—such as ads within utility apps or minigames—shows a different pattern. Engagement tends to concentrate toward the end of the ad, around the 25–30 second mark, when the 'X' (close) button UI appears and users are preparing to return to their original app. For this reason, Park emphasized the importance of creating assets that consider not just the content, but also the layout suited to the specific medium.

Last month, Appier participated in 'MAU Vegas,' a mobile app marketing conference in the U.S., where they held roundtables with leaders from various verticals. Park shared two key insights from these discussions.

First was the industry demand for an intuitive understanding of how and why creative engines work better on certain media. Second, as noted by one leader, was the realization that the core of a high-performing team is not better assets or campaign management, but an 'overwhelming performance measurement infrastructure.' This means that accurately measuring performance is just as important as creating high-quality assets.

Park introduced a system for deriving asset insights tailored to the age of AI agents. The approach involves having agents take over the repetitive processes performed by marketers—from asset planning and media setup to operations, report analysis, insight derivation, and secondary planning and design collaboration. Park described this as a five-stage circular structure.

In the 'Launch' phase, AI generates assets by combining historical industry data. Rather than using simple templates, the agent collects assets such as the brand's App Store/Google Play presence and website, then generates new assets that reflect the brand's tone and manner, guidelines, and custom rules. Park explained that while design teams could do this manually, the key is scalability and whether the agent can replace the back-and-forth communication between marketers and designers.

This is followed by a 'Winning Asset Selection' phase, where the performance of AI-generated assets (CTR, CVR, CPI, etc.) is measured to identify top performers, and a phase where real-time ad data is mapped to all assets. The core is the 'Inspect' phase. Here, a Vision Language Model (VLM) labels every element within video or image assets—such as whether they are live-action or animated, the use of zoom and text overlays, video length, and the presence of CTA buttons or app ratings. These decomposed elements are then combined with campaign performance data to identify which signals actually contributed to improved results.

Finally, a "Reason" phase is added, which goes beyond simple correlation data to explain why a specific element was or was not effective. Park noted that this directly addresses the demand from global marketers to understand why certain assets perform well. Based on these insights, the gap between "as-is' and "to-be" assets is identified, guiding the next round of production and closing the loop.

Based on research conducted by the Appier team over the past month or two, Park revealed that the element most strongly contributing to performance in the DSP ecosystem is the CTA (Call-to-Action) button. Since DSP users show peak engagement toward the end of an ad, as shown in the earlier graph, driving action at that moment is decisive.

Specifically, he suggested that CTA buttons should be continuously exposed and emphasized during the 25–50% segment of the video length; for a 30-second video, this corresponds to the 7.5 to 15-second mark.

Park concluded by summarizing that Appier has provided a framework for what is possible with Agentic AI, based on media-specific consumption strategies and the needs of global marketers. He ended his presentation by saying, "While marketers can apply these methods themselves, I hope you will explore them more deeply with Appier."