Introducing Professionals Series, which dives into the overwhelming expertise of professionals working within PayPay Group. This time, we spoke to data scientist Kentaro Minami about his noteworthy projects and the principles he values in data utilization.
Kentaro Minami
Leader of Merchant Credit Team, Finance Business Strategy Department, Finance Business Strategy Division, Finance Business Group
Earned a PhD in statistics. After working at an AI startup as a researcher focused on foundational machine learning and financial AI research and development, he joined PayPay in December 2023. Currently, he is involved in developing credit models for merchants and other related initiatives.
From the World of Research to PayPay
Could you share your career journey?
Originally, I specialized in systems within the faculty of engineering. However, my longstanding interest in theoretical fields like mathematics and statistics led me to switch my focus to statistics when data science began gaining popularity. I eventually earned a PhD in statistics and started my career as a corporate researcher at an AI startup.
The company I joined was involved in AI consulting, so I supported various companies in promoting AI as a researcher. It was through encounters with clients in the financial industry that I developed an interest in financial AI. In fact, the financial sector often employs advanced mathematical theories, making it a popular career choice for those who majored in physics or mathematics. I was drawn to the mathematical intrigue of the field and engaged in the research of pricing options—a type of derivative (financial product derived from common financial products such as stocks and bonds)—using deep learning, which I documented in papers and presented at international conferences, enjoying a fulfilling journey as a researcher.
What brought you to PayPay?
Despite achieving results as a researcher, I was troubled by the gap between academic accomplishments and the speed at which they materialize into products. I wanted to experience firsthand the process of a product growing as a business, from its research phase to reaching users. Therefore, I began job hunting, focusing on business companies in the financial sector.
During my job search, I heard about a credit project at PayPay leveraging machine learning, which piqued my interest. While considering joining a company where acquaintances from the research field were working, I chose PayPay for the new challenge it offered—a setting devoid of other data scientists.
Facing the Limits of Data as a Professional
What are your current responsibilities and mission?
Currently, I’m part of the Merchant Credit Team within the Financial Strategy Department, working on developing and improving credit models for merchant-facing financial services. Specifically, I am primarily involved in developing and improving predictive models for the screening process of the merchant service “PayPay Funding,” which I have been consistently participating in since joining the company. Utilizing the power of machine learning, I daily engage in activities to support PayPay’s financial services, contributing to the creation of innovative services that resolve merchants’ challenges.
What was an impressive aspect of your projects?
A particularly challenging and intriguing aspect was adopting a framework different from the mathematical models commonly used by Japanese financial institutions. For instance, in the field of credit modeling, which involves visualizing customer credit risk, the framework of calculating expected losses using PD/LGD/EAD (Probability of Default, Loss Given Default, Exposure at Default) is well-known.
However, PD/LGD/EAD framework is effective mostly when lending is involved. PayPay Funding employs a concept closer to RBF (Revenue-Based Financing)—a service where funds are provided based on future projected sales, and the principal is repaid alongside fees—which is observed in several foreign countries. The RBF world, being an emerging service, lacks established mathematical models, prompting us to develop a custom model predicting the repayment likelihood based on current sales. Consequently, we achieved a model that enables speedy, collateral-free, and guarantor-free funding, earning favorable responses from our users.
Naturally, working on a service like PayPay Funding involves more than just model research. Challenges like delivering releases by deadlines with limited staff and resources, and the difficulty of explaining complex concepts to internal stakeholders, were also present. Nonetheless, the excitement of developing unexplored models and the desire to deliver a service embodying my theories kept me motivated, leading to a successful release. By involving myself in user hearings and developing with user perspectives in mind, despite holding a data scientist role, we surpassed our expectations with over five times the anticipated applications.
Related Articles：Unleashing Five Times the Expected Results with Machine Learning Model and a User-First Approach: The Story Behind “PayPay Funding”
What do you consider important in your work at PayPay?
As a data scientist, I adhere to the mindset of “Be Sincere To be Professional.” Generally, data scientists are perceived to “uncover hidden secrets from data to solve problems,” akin to how data-based characters are portrayed in manga.
However, in a corporate environment, it’s crucial to recognize the things that cannot be derived from data and be aware of its limits, understanding that subjectivity inevitably intrudes when handling data. For instance, even if data is correctly processed, the growth rate induced from the same data can vary from 1% to 10% depending on definitions and interpretations.
Moreover, in such scenarios, the field usually anticipates high growth predictions. While it’s possible to deliberately extract only the desired data, artificial manipulation often results in significant deviations from reality during subsequent effect validations, ultimately leading to losses. As a data professional, I strive to always separate data from subjectivity and communicate while illustrating data’s limitations.
Fascinating Problems Abound at PayPay
From a data scientist’s perspective, what kind of company is PayPay?
PayPay is a company where the potential for business development utilizing data is expanding. We have access to vast amounts of data from 67 million users and merchants (as of December 2024), and we can analyze user behavior across diverse services including not just payments but also P2P, coupon usage, and financing. We possess not only a large volume of high-quality data but also data with high freshness, positioning us confidently among Japan’s leading data-holding enterprises.
Additionally, with a culture of “explaining everything through numbers” firmly established, it’s a conducive environment for data scientists. However, due to the high importance of data in decision-making, there can be an exaggerated expectation for data scientists to “provide numbers for everything” from those around. The high expectations necessitate a keen awareness of the data’s limitations.
What are your future goals and vision?
Moving forward, I aim to improve the predictive models and services to make the existing PayPay Funding service more user-friendly. Simultaneously, I wish to leverage the insights and knowledge gained from PayPay Funding to launch new services for merchants. I hope to deliver services manifested from my theories and passions to more people, while also aspiring to generate innovative cash flows through PayPay’s services. Through the company-wide goal of Financial Shift, I aim to contribute to establishing our position as a financial platform leader.
Could you leave a message for the readers?
Personally, I believe the appeal of a problem to solve influences a data scientist’s sense of fulfillment. In that regard, PayPay tackles services not yet realized by existing financial institutions, as well as unique issues arising from our innovative nature, presenting challenges without solutions in research institutions or academic papers. However, there’s an allure in unraveling unsolved mysteries, and an environment where the solutions we derive can bring happiness to numerous users. I invite everyone to join us in embracing new challenges together.
Here is the recruitment information
*Job openings and employee affiliations are current as of the time of the interview.