Among corporations' artificial intelligence (AI) projects, 60%–70% disappear after completing proof of concept (PoC) without being adopted for actual services. They remain in the lab because the results fall short of expectations or do not significantly help real work. Kim Hee-jun, chief technology officer (CTO) of Kakao Pay Insurance, met with ChosunBiz at the Pangyo office in Seongnam, Gyeonggi, on the 21st and said, "We should not run projects with the goal of adopting AI, but take an approach that solves pain points (the points where customers feel inconvenience) with AI."
The "overseas medical expense AI claims review service" developed by Kakao Pay Insurance is the result of building to solve employees' and customers' pain points. Domestic medical expense claims have standardized formats for receipts and itemized statements, and diagnoses are organized by codes, but documents issued overseas differ in everything from institution names and medical certificate terminology to types of prescribed drugs. Most are also written in foreign languages. Because of this, reviewers had to verify one by one whether the hospital actually exists and whether the prescribed drugs were legitimately prescribed. Even a skilled claims expert had to spend as little as 40 minutes to more than two hours per case. Kakao Pay Insurance used Kiro, Amazon Web Services (AWS)' AI development tool, to build the related system and cut this review to 2–3 minutes.
Kim, the CTO, said, "What customers want most is speedy insurance payouts," adding, "To achieve that, we had to solve the issue of the long time and resources required for claims review." The following is a Q&A.
─ Why did you move to develop the overseas medical expense AI claims review service?
"What customers ultimately want most is speedy insurance payouts. To achieve that, we needed an automated payment system, and we had to solve the issue of the long time and resources required for claims review. Domestic medical expense claim reviews have a certain degree of standardization in terms, diagnosis codes, and documents, so the effort required from the reviewer can be relatively small. In contrast, overseas medical expense claim reviews differ greatly from domestic circumstances in the types of documents, the terms used for diagnoses, and the diagnosing institutions. Work is needed to compare, understand, and convert each item. Depending on the situation, even a skilled claims expert needs at least 40 minutes to more than two hours for this work.
Kakao Pay Insurance sought to improve this work through AI. A claims expert still makes the final decision, but we focused on drastically reducing the effort required to reach that decision."
─ How accurate is it?
"AI hallucinations mostly occur during reasoning, but financial firms' AI uses retrieval-augmented generation (RAG), which derives results based on facts rather than reasoning. Because it keeps substantiating the grounds for its answers, the accuracy is quite high. Of course there are false positives, but people verify and reference them, so it is different from leaving everything to AI."
─ Were there any difficulties during development?
"As AI has exploded, many companies run PoC projects, but several studies show that only 20%–30% lead to actual services. Some 70%–80% fail. That's because approaching it as a technology task often ends up detached from real work. From the perspective of front-line departments, they think, 'They're doing yet another thing,' rather than believing it will help their work.
In this project, we personally promised it would be "a weapon to reduce your overtime and solve real pain points," and the on-site response was, "Then can we also give lots of input?" We built consensus to make the journey of building the service together. The most important part was not "let's adopt AI," but an approach of "let's use technology to solve what is hardest for us right now."
Kakao Pay Insurance runs projects with several must-keep principles for PoC, especially AI technology PoC. They are clarifying the problem to be solved, not doing PoC for technology's sake; specifying the constraints of the project; and, lastly, using currently available technologies."
─ You built this service with AWS Kiro.
"It was important to find a trustworthy partner that best fits our development culture and corporate culture, and that was AWS. AX (AI transformation) is centered on a Cloud-Native culture, and AWS was the cloud service that best supports that culture. AWS Kiro is the AI integrated development environment Kakao Pay Insurance is using most actively. Kakao Pay Insurance uses Kiro not only for developers' coding, but also for planners' and front-line departments' work. Working by spec, not by prompts, suits our culture well. We received great help across the process from eliciting and analyzing requirements through development."
─ When can customers start using this service?
"For a financial company to adopt AI in an actual service, it must go through the innovative financial service (financial regulatory sandbox) program. We are considering applying within the year."
─ Do you plan to expand this service to other areas?
"Our focus was not 'AI-enabled claims review,' but 'solving claims review pain points with AI.' If the same pain points exist in other types of claims, we can solve them with this technology. Because we developed it in-house without external solutions, it is advantageous for expansion to other services."
─ What other AI services has Kakao Pay Insurance developed?
"A representative example is the instant insurance payout service using document recognition. Our in-house AI document recognition model identifies the documents customers submit to file a claim and then extracts key data. It has been applied to mobile phone insurance claims review, and there was a case where the payout was made about 18 seconds after the claim was filed. It has also been applied to flight delay compensation for overseas travel insurance, with a case where a customer received the payout in just one second."
─ Do you think AI has reached a stage where it meaningfully affects the company's profit and loss?
"I think it is still early to view AI from the perspective of expense. The right expression is not 'AI replaced people and reduced costs,' but 'thanks to AI, people are doing more valuable work.' Shortening what used to be a long insurance payout experience and positively changing customers' perception of insurance is, in my view, a far more meaningful impact than an arithmetic calculation that expenses were reduced."
─ As CTO, how do you use AI in daily work?
"We use AI not just in code, but across various areas such as defining work, planning, and requirements analysis. We use AI agents that analyze and review code written by developers. We also use it to summarize documents or extract questions. In addition, Kakao Pay Insurance does not have a separate AX (AI transformation) organization, but almost every department uses AI in various ways. It is actively used at each stage—planning, development, design, and corporate planning."
─ Any final comments to add?
"Big Tech AI firms are rushing to claim 'our model is the most powerful,' but using that model does not make our service the most powerful. The key is to use AI where customers want it most and solve that problem. In other words, the focus is not which model you used, but where you apply AI to draw out customer value. We live in an era when today's most powerful model becomes outdated in a few months. Corporations should focus on securing models and technologies specialized for processing their own data and domains.
Kakao Pay Insurance, as a financial company, places importance on the responsibilities borne by financial contracts. We therefore judge that securing trustworthy AI technology is most important. These include technologies such as RAG, which reduces hallucinations and produces answers with clear grounds; guardrails for stability; and caching, which lowers expense and speeds up responses."