Innovating the Advertising Business with Technology: The Challenge of Prism Partner

― What were the reasons for choosing CyberAgent as a partner, and what strengths were you initially expecting?

Tanazawa: CyberAgent consistently holds a top-tier share of ad sales across major platforms such as Google, Meta, LINE, and Yahoo! Beyond their sales power as an ad agency, we were drawn to their deep understanding of 'what makes a product sell.' They were also early movers in the digital promotions field, which aligned with our goal of expanding the digital promotions market by leveraging DOCOMO's assets, such as d POINTs and d Payment. With strong support from DOCOMO's management, this alignment led to the establishment of the company.

― Were other potential partners in the running?

Tanazawa: DOCOMO is always exploring collaborations across many different fields. In the realm of internet advertising, partnering with the number one player, CyberAgent, was a critical move for us. To fully grow our advertising business, their sales power, technological prowess, and sheer industry presence were compelling.

― What technological developments have you undertaken for the "Prism Partner DSP"?

Katsuragawa: There are two main developments. The first is a creative selection feature. This function maximizes delivery effectiveness by automatically identifying and utilizing the optimal creative. While A/B testing has traditionally been the standard method, it is costly to ensure high accuracy because it requires equal impression allocation. Therefore, we focused on developing a more efficient alternative.

Kaneko: We utilized a bandit algorithm, a field within reinforcement learning. Specifically, we adopted a relatively new method called Top-Two Thompson Sampling (TTTS)*. This innovation allows us to automatically identify the optimal creative with high accuracy and minimal cost, thereby maximizing advertising effectiveness.

*Top-Two Thompson Sampling (TTTS): A reinforcement learning technique that uses a probabilistic approach to efficiently find the best option. It's especially powerful for situations like ad creative optimization, where you need to make effective choices while minimizing the number of trials.

Katsuragawa: On top of that, our paper on this feature was accepted into the Industry Day track at WSDM 2025, one of the top international conferences for web research. Industry Day is where the world's leading tech companies come to share ideas. I believe we were recognized for two things: the novelty of applying a new method, like TTTS, to creative selection in online ads, and the practicality of demonstrating that it works in a physical world.

Kaneko: Even at CyberAgent, it is not common for research implemented in actual services to be accepted as a paper. The acceptance of technological contributions from a collaborative project is particularly challenging compared to those from in-house research. This case was the first achievement for us made from a joint venture and was successfully integrated into a service.

Katsuragawa: The knowledge and experience we have within CyberAgent were very helpful in deciding to adopt this method. The collaboration with the members of AI Lab, CyberAgent's AI research wing, who co-authored the paper with us, was absolutely essential.  

― So this was a true collaboration between the business and research teams to bring it to life. Could you tell us about the other key achievement?

Katsuragawa: Another core feature we've built into the Prism Partner DSP is called Active Insight. We developed it by leveraging DOCOMO's data with machine learning. Its purpose is to optimize ad delivery by identifying users with high purchase intent based on their behavior and then targeting them.

Kanbe: Using Active Insight, we've seen cases, for example, where we drove traffic from the Prism Partner DSP to Amazon store pages and achieved a return on ad spend (ROAS) of over 400%. We're also seeing great results from our promotional campaigns in the retail sector that leverage the d POINT ecosystem, boosting sales for manufacturers and retailers.

Mr. Tanazawa: It's a perfect example of strong technological skill of our data science team—the ability that gets papers accepted at international conferences—has directly produced concrete business results like a ROAS of over 400%.

― I heard the timeline was incredibly fast—the concept was solidified in April, and implementation was finished by July. How did you design your decision-making process to make that happen?

Kanbe: CyberAgent is known for its rapid decision-making, whereas DOCOMO, as Japan's largest telecommunications carrier, must be particularly cautious with data-related judgments. Mr. Tanazawa's skill in balancing these two approaches is a key strength of Prism Partner. As a startup, we need to develop products quickly. This decision-making structure, which balances caution and speed, is playing a vital role in our growth. 

Tanazawa: As part of DOCOMO, we respect the need for a careful, multi-faceted decision-making process. But within Prism Partner, we've built a structure that enables much faster decision-making. I believe this is what underpins our business speed.

― Lastly, what challenges do you plan to take on in your future business development？

Tanazawa: We want to continue creating new ad products by combining DOCOMO's assets with CyberAgent's technology. Specifically, we're looking to enhance our integration with new data sources, such as ID-POS data, to deliver a truly unified online and offline marketing experience.

Kanbe: We believe our mission is to provide valuable solutions for advertisers by maximizing the use of DOCOMO's data. The use of retail media and data holds significant potential for future markets. We leverage our AI technology and data science expertise to deliver higher-precision marketing solutions, including advanced advertising products.