South Korean telecom operator KT won the project to rebuild the chatbot and advisory bot of Woori Bank, one of South Korea's largest banks, with a concrete goal: get AI agents to stop merely answering questions and start completing banking transactions from start to finish.
The centerpiece of the project is "Agent Connector," a KT solution that links different channels and services — the chatbot, the advisory bot, the "AI banker" service, and task agents — while preserving conversation context across all of them. If a customer starts an inquiry over chat and then switches channels, the service can pick up exactly where it left off, without the customer having to repeat their situation. That shared context is what lets an AI agent, not just a human, take over the conversation and actually execute the task the customer requested.
KT describes the goal as building an "agentic AICC" (AI-agent-powered customer contact center) environment capable of processing real tasks that come out of an inquiry, not just providing information. In practice, this moves Woori Bank from a chatbot that routes customers toward a human channel, to one where the AI agent itself executes the requested banking action — while keeping the conversation thread intact across channels.
The Woori Bank case represents a transition also being discussed in Latin American banking: moving from conversational bots that resolve FAQs to agents with permission to execute actions inside core banking systems (transfers, data updates, product requests). That's a bigger leap in risk than an informational chatbot, because an agent error is no longer just a wrong answer — it's a mis-executed transaction.
If your company operates in financial services or any regulated sector in Guatemala or Latin America and is evaluating the jump from informational chatbot to transactional agent, the Woori Bank case works as both a roadmap and a warning at the same time: the customer-experience gain (not repeating your request when you switch channels, resolving the transaction on the first interaction) is real, but it requires investing first in the context-continuity layer and permission controls, before exposing the agent to executing actions involving real money. Starting with a scoped pilot (say, a single low-risk transaction type) before scaling to the full service catalog is the difference between a controlled migration and a costly incident.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel