Validating the Agentic AI revenue model that scales GPU and API cost constraints
[소셜밸류=박완규 기자] On the 29th, Korean AI startup AICLUDE announced that it signed a joint research agreement regarding Avatar-based Agentic AI with STNet, an ICT subsidiary of the Shikoku Electric Power Group, to conduct in-store validation of avatar-based Agentic AI across Japanese stores.
AICLUDE achieved a remarkable result by building trust through direct engagement with Japanese firms, following its participation in SusHi Tech Tokyo through the Seoul Startup Hub Gongdeok Global Expansion Program.
Agentic AI refers to next-generation artificial intelligence that sets goals and makes decision without human intervention, automatically completing complex tasks through tools and learning processes.
The project gained traction as it highlighted real-world issues regarding a low return on investment relative to high operational costs with traditional agentic AI services which rely on GPU computing and external API calls.
With labor shortages and an aging population in the Japan retail market, the demand for management efficiency is growing significantly, along with rising interest in AI-based automation services. In fact, costs of GPU infrastructure and external API usages posed a burden for small and medium sized stores and multi-unit retailers, as these resources were essential to maintain steady, high-quality generative AI services.
Under these circumstances, STNet focused on establishing a sustainable revenue model rather than simply conducting technical trials of generative AI.
Following a discussion with STNet, AICLUDE entered into a joint research agreement on agentic AI avatar-based retail. The joint research is being conducted at two Pikara Shops, a STNet-affiliated telecommunications store located in Takamatsu, Kagawa, Prefecture, in the Shikoku region of Japan.
The research validation period runs from last December through this coming March and is designed not as a short-term technological demonstration, but as a phase focused on data accumulation and analysis through repeated operations.
Based on its proprietary agentic AI technology, AICLUDE highlighted inefficiencies of conventional methods that repeatedly invoke generation processes for avatar expression and dialogues. Thus, it minimized unnecessary GPU computation and external API calls by separating video generations for avatar movement from dialogue processes and by optimizing the system architecture to prevent the repeated generation of identical or similar outputs.
The PoC targeted passengers passing by the stores with AI avatar attracting them into the stores and introducing services and items through conversation.
As AICLUDE’s agentic AI system is integrated with the operations management platform, managers at headquarters and in stores can monitor key performance indicators in real time, such as operating hours, visitor characteristics, store environmental changes and more. Then, it validates AI avatar models that evolve according to on-site operational conditions by continuously adjusting and improving the AI agent’s decision-making logic and dialogue scenarios based on collected operational statistics.
Moving beyond just a short-term PoC, the joint research also evaluates the feasibility of future commercialization, including integration with STNet’s proprietary cloud system. Based on PoC results, both companies aim to develop an AI avatar service model optimized for the Japanese retail market and to discuss business expansion for multi-store adoption and operational efficiencies.
Kim Keun-jin, the CEO of AICLUDE, stated “Throughout the project, we aim to overcome the burden of agentic AI avatar technology and establish a benchmark demonstrating its practical applicability in foreign retail stores, and pave the way for local exports.”
Meanwhile, AICLUDE plans to assess the potential of global expansion of Generative AI and generative AI technologies through PoC in Japan, and to promote commercialization in global market including Japan in phases.
[ⓒ 사회가치 공유 언론-소셜밸류. 무단전재-재배포 금지]
