MagicVoice Enhances Traditional ARS with Direct Natural Language Answers from sLLM Callbot
Overcoming Limitations of Previous ARS Systems

Transforming Tens of Billions Won AICC Projects into a Cost-Efficient Solution
Unified Intelligence Core Enables One-Stop Callbot, Chatbot, and Consultation Chat on a Single GPU

Image provided by Makebot

Image provided by Makebot

View original image

Makebot Co., Ltd., a conversational AI specialist led by CEO Kim Jiwoong, announced on June 15 that it has successfully built and launched a next-generation AICC (AI Contact Center) callbot system at the customer touchpoint of the Korea Federation of Savings Banks (KFSB), which supports all savings banks nationwide. This marks the opening of a new era of fully unmanned customer service in the financial sector and maximizes customer response efficiency.


The Korea Federation of Savings Banks is the central institution operating the nationwide joint IT system and integrated digital banking platform ‘SB Toktok (SB Toktok Plus)’ for savings banks, providing integrated customer service for approximately 80 member savings banks. The new AICC callbot has been applied to this integrated customer touchpoint, laying the foundation for expansion to all member banks through KFSB’s standard specifications. Notably, the KFSB and Makebot jointly established a standard AICC callbot model optimized for the financial sector through this project, providing a foundation for member banks to easily adopt and expand the solution in the future.


The introduction of the KFSB callbot was designed by Makebot based on its independent analysis of the entire infrastructure of both primary and secondary financial sectors in Korea (including savings banks, capital companies, and insurance firms). The company explained that this is the first case in Korea to demonstrate “24/7 fully automated callbot-based service” across both primary and secondary financial institutions. From the moment the system went live, it has been handling hundreds of daily deposit and loan inquiry calls via natural language conversations without human intervention, significantly reducing labor and operational costs for financial institutions.


Previously, the “visual ARS” systems adopted by many financial companies simply displayed the menu structure of a voice ARS on a screen. Customers had to manually navigate and select each item step by step, making usage inconvenient, and, most importantly, the AI could not directly answer customers’ actual questions—a fundamental limitation. Makebot’s AICC callbot solution, “MagicVoice,” does not just replace visual ARS; it supplements it with a generative AI consultation layer that provides immediate answers to natural language questions, thereby overcoming the limitations of traditional ARS.


For example, when a customer calls the representative number “1544-3637” for SB Toktok (the savings bank digital banking app), they can choose between visual ARS or regular voice ARS. After this, if they ask questions like “How do I register a certificate?”, “How do I register an OTP?”, or “How do I repay a loan?” in natural language, the callbot responds immediately. Makebot’s advanced natural language understanding (NLU)-based AI callbot can interpret the customer’s intent and handle inquiries in a single conversation without menu navigation.


In particular, for financial product information that fluctuates frequently, such as interest rates, the callbot searches the core banking system in real time and provides the latest information via voice. Unlike traditional ARS, which replayed only preset answers, this system delivers accurate, up-to-date information by reflecting real-time data—a key differentiator.


Above all, MagicVoice is designed so the AI responds in natural language, but if the customer wants to speak to a human agent, the system immediately switches to a human advisor. The callbot handles simple, repetitive inquiries 24/7, while only complex or sensitive consultations are transferred to human agents, greatly reducing the workload per agent and improving response efficiency. This represents a reallocation of work, allowing both AI and human agents to focus on their respective strengths.


The technical achievement is also noteworthy. Makebot directly installed high-performance GPUs and a compact large language model (sLLM) compliant with security regulations within the KFSB’s internal infrastructure. In particular, the company applied, for the first time in Korea, a “hybrid inference architecture” that organically combines its proprietary deep learning-based natural language processing (NLP) engine and sLLM, thereby meeting the strict accuracy requirements unique to the financial sector and increasing the reliability of answers. Makebot plans to use the standardization within the KFSB as a bridgehead to quickly expand this AI agent infrastructure not only to all member savings banks, but also to the wider secondary financial sector—including capital, credit card, and insurance companies—as well as public institutions and large enterprises across various industries.


This launch is also significant for breaking the cost bottleneck that has existed in the AICC market. Previously, financial companies wishing to build an AICC had to spend tens of billions of won purchasing solutions through large SI consortiums, or implement callbots, chatbots, and consultation chat systems separately. In recent years, companies also faced the burden of buying expensive GPU servers for each project. This resulted in development periods of over a year, creating high entry barriers for mid-sized financial institutions.


In contrast, Makebot runs the entire AI infrastructure stack—including the AICC callbot “MagicVoice,” AI consultation chat “MagicTalk,” scenario-based chatbots, and intelligent knowledge search engines—on a single GPU server, the “Unified Intelligence Core.” Unlike previous methods where hardware and software fragmentation led to redundant costs, this approach allows the implementation of callbot, chatbot, and consultation chat all on one core, without needing to purchase separate solutions or additional GPUs for every project. As a result, Makebot has drastically shortened the deployment period, enabling the introduction of high-quality AICC systems with a reasonable budget of just hundreds of millions of won and achieving an overwhelming return on investment (ROI). Industry observers note that Makebot is virtually the only company capable of providing a full stack of AICC components in a one-stop solution at the cost of a competitor’s single product.


Makebot’s cost-efficient AICC model, based on the “Unified Intelligence Core,” is attracting industry attention for dramatically lowering the AI adoption barrier for small and mid-sized financial institutions. Especially since the standard specification enables the rapid deployment of full-stack AI consultation infrastructure without the need for individually building expensive systems, intelligent contact centers are expected to become popular not only in the secondary financial sector, but also across all industries previously hesitant to adopt due to budget constraints.


In fact, Makebot is rapidly expanding its presence in the financial and enterprise markets. Recently, it successfully implemented the next-generation AI-based consultation chat solution “MagicTalk” at Woori Financial Capital. MagicTalk operates as an “AI-collaborative” system, where a large language model instantly generates an optimal draft response while an agent consults with a customer via real-time chat, and the agent reviews and finalizes the response. It is recognized as the first case in Korea where an sLLM has been deployed within the strictly regulated internal networks of the financial sector and commercialized for consultation work. Companies that have adopted the solution have seen average consultation processing times reduced by more than 30%. Makebot has officially released the standardized “MagicTalk SaaS” version and is expanding its adoption to major Korean companies such as Samsung Electronics.


Kim Jiwoong, CEO of Makebot, stated, “The achievement of launching the nation’s first 24/7 fully unmanned financial callbot across both primary and secondary financial sectors was possible thanks to the bold digital vision and unwavering trust of the leadership and staff of the Korea Federation of Savings Banks, who oversee the mutual prosperity and digital advancement of over 80 savings banks nationwide.” He further emphasized, “This hybrid architecture, which we developed after much deliberation with the federation, will serve as a signal for fundamentally transforming the existing high-cost, supplier-driven AICC structure.”


Kim added, “Based on our patented technology and the single GPU Unified Intelligence Core, we have successfully met financial compliance requirements while drastically lowering implementation costs. We plan to establish this success as a safe and reliable ‘AICC standard model’ across the entire secondary financial sector, and use it as a bridgehead to expand AI agent infrastructure from enterprise to the global SaaS market.”



Meanwhile, Makebot is an AI specialist with the most extensive experience in building AI chatbots in Korea, ranging from NLP to LLM and generative AI. It has supplied chatbot services to major companies and public institutions—including KT, CJ, Korean Air, OnStyle, Lotte, Korea Electric Power Corporation, and Korea Tourism Organization—as well as hundreds of hospitals including Gangnam Severance Hospital. Its proprietary RAG (Retrieval-Augmented Generation), hallucination removal technology, and “HybridRAG” have been recognized for their technological excellence at SIGIR 2025, the world’s most prestigious AI academic conference. The company plans to lead digital innovation across finance, public, healthcare, aviation, distribution, and other industries with its integrated AI infrastructure—including the AI callbot MagicVoice, AI consultation chat MagicTalk, chatbots, and knowledge search engines.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing