빅데이터 분석과 LLM 모델을 이용한 고객 페르소나 시나리오 개발 및 마케팅 활용 방안에 관한 연구

This study presents a novel methodology for persona development by combining quantitative methods of Data-Driven Persona Development (DDPD) with qualitative enrichment through Large Language Models (LLMs). Customer personas are widely recognized as essential tools for understanding customers more deeply and delivering personalized experiences. Traditional persona creation, however, has relied heavily on small-scale interviews and surveys, resulting in limitations in representativeness, reproducibility, and adaptability to rapidly changing markets. In contrast, DDPD offers advantages in terms of reliability, efficiency, and scalability by leveraging objective analyses of large datasets. This study extends the DDPD framework by demonstrating that even in contexts where firms lack proprietary customer data, such as startups, personas with narrative depth and richness can still be developed through the integration of public datasets and LLM-based models, and subsequently applied to strategic marketing. For this research, a dataset consisting of 5,760 traveler records and 36 variables was constructed using Jeju travel log data provided by AI-HUB. Principal Component Analysis (PCA) were applied in conjunction with K-means clustering, resulting in the identification of four distinct clusters that varied in demographic profiles and travel patterns. Building on these quantitative findings, a Custom-GPT for persona development was designed to generate persona skeletons automatically and produce customer journey–based scenarios linked with POE (Paid-Owned-Earned) media strategies. By doing so, the model moved beyond simple data summarization to deliver actionable strategic insights. Importantly, the Custom-GPT was programmed to incorporate academic theories and established frameworks, ensuring that the generated scenarios were grounded in theoretical rigor while also offering practical utility. The results highlight the potential of hybrid persona development methods that merge statistical clustering with generative AI narrative generation. The proposed approach contributes theoretically by expanding DDPD research to encompass contexts with limited proprietary data, and practically by providing a scalable and adaptable framework for startups and new business initiatives. Ultimately, this study underscores the transformative potential of integrating big data analytics with LLM-based models, paving the way for more dynamic, reproducible, and strategy-oriented persona scenarios that can support marketing and policy decision-making in data-rich and data-scarce environments alike.