[Edaily Reporter Lee So-Hyun ] Kakao(035720) To terminate the contract with a single seller who had ceased activity on the “Send as Gift” service, a staff member had to gather data from multiple systems, determine whether to terminate the contract, and then send a certified letter. Switching back and forth between Google Sheets, documents, and a separate Windows PC to complete this process took about one hour per seller.

Kakao redesigned this process using artificial intelligence (AI) and automation technology, reducing the processing time per case from one hour to 10 minutes. It is a “semi-autopilot” approach, where human approval is required at specific stages rather than full automation.

On the 14th, Kakao shared this case study at “if(kakao)26” and introduced the role of the FDE organization, which was newly established to drive the company’s AI transformation (AX). FDE is not an organization that centrally develops general-purpose AI tools; rather, it is an organization that goes directly into the field to identify workflow bottlenecks and collaboratively transform data, processes, and operational methods.

Kim Jeong-han, a Kakao representative, noted, “The sum of individual productivity does not always equal the sum of organizational productivity.” He explained that while introducing AI tools may speed up individual tasks such as document creation or searching, the waiting times for review, approval, and handoffs between people and teams remain unchanged. He described this as “the dots have improved, but the lines connecting the dots remain the same.”

FDE actively identifies these bottlenecks. When selecting projects, they examine: △ whether key metrics—such as costs, revenue, and risk—are actually changing; △ whether there are many repetitive tasks or frequent errors; and △ whether improving one area will speed up subsequent processes. Their goal is problem-solving, not feature development; rather than receiving planning requests, FDE defines problems directly on-site.

A prime example is the process for terminating contracts with gift card sellers. Previously, data on sellers, products, and orders was scattered across multiple locations, and the criteria for determining contract termination relied on the experience of the person in charge. Sending certified letters had to be handled on a separate Windows PC.

Kakao consolidated the data and shifted to a “report → assess → execute” workflow. Approximately 80% of decisions are now handled through statistics, trend analysis, thresholds, and business rules, with human approval required only at points where it is absolutely necessary. For sending certified letters through the post office—which lacks an API—Kakao applied RPA, which is well-suited for structured tasks, instead of LLM-based computer automation. Manager Kim emphasized, “Expensive models aren’t necessarily required to solve all these problems.”

Kakao’s goal is to go beyond automating individual tasks and expand this into a platform. The vision is to develop FDE—which bundles data connections, decision-making logic, workflows, and AI agents created in the field—into an “AI Workspace as a Service” that can be reused by other organizations.

“Sending FDE to the field is just the beginning,” said Manager Kim. “The real question is whether the results will remain as a reusable platform.”