As companies accelerate their AI transformation (AX) initiatives, many that have already adopted AI are expressing frustration that their AI chatbots generate irrelevant, inconsistent, or unreliable responses. Industry experts argue that the root cause of this low level of satisfaction often lies in the lack of a well-designed taxonomy--a structured way of organizing data so that AI systems can correctly interpret and utilize it.
Kwang-deok Seo, Country Manager of ThinkingAI Korea, said, “Many companies attribute AI project failures to the performance of the AI model itself. In reality, however, the problem is usually the data.” He identified the absence of a proper data classification system as one of the leading causes of unsuccessful AI projects.
“Even when the purchasing behavior appears similar, there is a significant difference between a first-time payment, a subscription renewal, and an additional purchase,” Seo explained. “Humans naturally distinguish these through context, but AI interprets information literally. Unless data naming conventions and storage structures are defined in advance, the data cannot be effectively utilized.”
This is where taxonomy plays a critical role. Taxonomy, which literally means a classification system, refers to the practice of systematically organizing information according to defined rules. While taxonomy has traditionally been used in data analytics and marketing, its importance has grown significantly with the rise of AI.
Seo noted that companies commonly encounter three recurring challenges when building a taxonomy. First, even after establishing the overall design direction, organizations often struggle with the detailed design decisions, such as how to define individual events and which elements should be treated as attributes. Second, the taxonomy must be designed with scalability in mind so that new data can be incorporated as services evolve. Finally, comprehensive documentation is essential.
“When employees leave the company or move to another department, those who remain are often forced to guess what a particular event was originally intended to represent,” Seo said. “In many organizations, taxonomies created without proper documentation continue to be used simply out of habit.”
To address these challenges, ThinkingAI incorporates AI into the data design process from the very beginning. AI first performs repetitive tasks such as data structuring and documentation, while the company's Customer Success team validates key performance indicators (KPIs) and business context to complete and refine the design.
“In the past, taxonomy design typically required anywhere from several days to several weeks,” Seo said. “Today, by collaborating with AI agents throughout the taxonomy design process, we can complete the work within just a few hours.”
He added, “Without the expertise accumulated over the past decade of operating a data analytics platform, it would not have been possible to entrust this work to AI. By enabling AI to participate from taxonomy design through data governance implementation, we aim to help enterprises adopt AI more quickly and efficiently.”
