[DigitalToday reporter Chi-gyu Hwang] "AI competitiveness comes not from models but from data that agents can trust and act on."

Lee Kyung-jong (이경종), head of KB Kookmin Bank and KB Financial Group’s Financial AI2 Center, said in a keynote speech at the Snowflake World Tour Seoul held at COEX on Aug. 27 that as model performance has become standardized, real differentiation comes from data quality and how it is used. He summed up KB Financial Group’s agentic AI strategy as AI that acts beyond search. He said to fully realize AI’s impact, companies need agentic AI that acts on its own when needed based on internal data, beyond AI that simply answers questions.

Lee said most AI used by companies is search AI that answers questions and provides documents and reports. He said the approach is useful but has limits. He said employees still need to recheck results and enter them directly into systems. He said the direction is AI that acts.

According to Lee, acting AI automates various operational tasks that people had planned and completes complex multi-step processes by planning them itself. He said this allows employees to focus on important work and raises productivity.

Building acting AI is not easy. It requires substantial preparatory work.

Lee presented three requirements for building acting AI: a foundation model in which reasoning and function calling work naturally; providing agents with the means to perform actual work by connecting multiple systems through an MSA-based system and MCP; and establishing a structure in which the data that serves as the basis for actions can be trusted. He particularly stressed building a data environment for AI. He said a data environment that AI can understand is needed to build AI that understands business context and acts.

Lee said data-related problems faced by companies seeking to adopt AI can be summarized into three main areas.

First is a data layer issue, meaning data is not accumulated in a way AI can understand. Lee said current data is accumulated around transaction processing and batch jobs, and a large portion of structured and unstructured documents are not properly structured. He said because business context is not reflected, AI cannot properly grasp the meaning even if it receives the data.

Next is platform connectivity. Lee said systems and platforms are scattered, requiring separate development work each time new data is connected. He said speed falls if the same work is repeated at each service application stage.

Last is an infrastructure issue. Lee said existing IT infrastructure struggles to keep up with the pace of change in AI technology, and scalability needed for real-time processing and large-scale computation is limited. He said in such circumstances, even if agents use good models, it is difficult to establish the data foundation needed for them to act in practice.

KB Financial Group is moving to build an environment for acting AI around two pillars.

One is changing data into a meaning-centered structure. Lee said the key is converting data beyond simple storage into data containing business context and meaning, so that when AI receives the data it can immediately grasp what it means.

The other is an on-demand structure that can connect immediately when needed. To do that, KB Financial Group introduced the Snowflake platform and changed its data operations. Lee said it shifted away from a model managed unilaterally by the IT department to a structure in which business departments that know the data best directly take ownership and produce and manage it. He said when business departments produce domain data, they register it in the Snowflake platform data marketplace, and employees and AI can directly pull and use needed data from the marketplace.

KB Financial Group began building a dedicated platform for agent AI development in 2024 and later converted it into an official service. In 2025 it first introduced a front-office agent supporting wealth management and retail finance, then rolled out a wealth management agent and a financial consultation agent in succession.

The wealth management agent helps analyze customer portfolios and quickly identify market issues, while the financial consultation agent helps with consultation work across multiple sales counters. KB Financial Group plans to continue increasing the number of AI agents it provides.

This year, KB Financial Group is expanding agent development into mid-office and back-office areas.

Lee said a risk management agent checks market and credit risks in real time and automatically detects anomalies. He said a loan screening agent automates the loan review process to improve speed and accuracy. He said a legal support agent is being developed with the goal of automating contract review, monitoring regulatory changes and compliance checks.

KB Financial Group has also deployed AI agents across the full software development process, building an environment that supports planning, development, testing and deployment. A data analysis agent covers data aggregation and preprocessing, comprehensive analysis and report writing.

Lee shared three points about the future of the agentic AI strategy.

First, he plans to integrate KB’s data assets based on Snowflake to complete a structure that agents can use immediately, and to expand the scale of agents across all front-office and back-office areas.

He also aims to use MCP-based system connections to move beyond simple answers and transition to acting AI that directly manages tasks. Lee stressed that the AI data system KB Financial Group is building is a long-term competitive strategy that rivals cannot easily copy.