Credit Scoring of Retail Customers in Neobanks: Developing a Hybrid Model Using Content Analysis and Multi-Criteria Decision-Making

Purpose: The primary objective of this study is to develop a robust and context-specific credit scoring model tailored for retail customers in neobanks, leveraging both traditional and non-traditional data sources. Considering the limitations inherent in conventional credit assessment frameworks, particularly for individuals lacking formal credit history, this research seeks to bridge the gap by integrating behavioral, legal, digital, and socio-demographic indicators into a comprehensive, data-driven model. The study adopts a mixed-method approach to first identify and classify the most relevant credit-related attributes through the qualitative content analysis of expert interviews. Subsequently, the Best-Worst Method (BWM), a multi-criteria decision-making technique, is employed to prioritize and assign relative weights to these attributes. The model aims to enhance the precision, inclusiveness, and efficiency of credit risk evaluation processes in neobanks, which operate in a fully digital environment and have access to real-time, high-volume user data. By focusing on retail customers, particularly those underserved or excluded from traditional financial systems, the proposed model contributes to improving financial inclusion and informed credit decision-making. The study not only offers theoretical contributions to the domain of fintech and risk modeling but also provides actionable insights for practitioners who develop AI-driven credit assessment tools in digital banking ecosystems.
 
Methodology: This study adopts a mixed-methods approach, combining qualitative and quantitative strategies to design and validate a comprehensive credit scoring model for retail customers in neobanks. The research is structured in two sequential phases, following the methodological framework of Saunders et al. (2016). In the first phase, qualitative data were collected through semi-structured interviews with 22 experts in digital banking and fintech. These interviews were analyzed using conventional content analysis (Hsieh & Shannon, 2005), employing open coding techniques to extract the key indicators influencing creditworthiness. The resulting codes were then grouped into thematic categories, ensuring data saturation and conceptual clarity (Glaser & Strauss, 1967).
In the second phase, the identified indicators were evaluated and prioritized using the Best-Worst Method (BWM), a robust multi-criteria decision-making tool introduced by Rezaei (2015). This method enables the systematic comparison of indicators based on expert preferences, while minimizing inconsistency and data redundancy. The integration of qualitative insights and quantitative weighting ensures that the proposed model is both empirically grounded and contextually relevant. This methodological design allows for a nuanced understanding of credit risk in a digital context, capturing both traditional financial behaviors and emerging digital signals relevant to neobank operations.
 
Findings and Discussion: This study presents a systematic and empirically grounded model for retail credit scoring in the context of neobanks, integrating traditional financial indicators with alternative digital and behavioral data. The findings, derived through a two-phase mixed-methods approach, reveal a hierarchy of influential credit risk factors, prioritized with the Best-Worst Method (BWM) based on expert judgment.
The most significant determinant identified is financial transactions and behavioral patterns, which received the highest weight (0.360). This dimension includes metrics such as payment history, spending behavior, savings consistency, and financial engagement across banking platforms. These variables offer a reliable proxy for financial stability and are readily accessible through the digital infrastructure of neobanks.
Ranked second, the legal and insurance profile (0.279) encompasses indicators such as criminal records, legal compliance, and insurance participation. These elements reflect the individual’s commitment to formal institutional systems and their ability to manage unforeseen risks, namely the factors closely tied to credit reliability.
The digital activity and online behavior category ranks third (0.200), underscoring the growing significance of alternative data in digital banking. Variables such as online shopping patterns, social media behavior, and app usage frequency serve as behavioral proxies for risk tolerance and consumption consistency.
Demographic and household data (0.120), including marital status, education, employment history, and residence stability, were found to exert moderate influence. Finally, risk-taking and philanthropic behavior (0.039), while less impactful in isolation, provide complementary insight into personal values and crisis-time financial conduct.
These findings highlight the value of integrating multi-source data for more nuanced and inclusive credit evaluations. The proposed model enables neobanks to assess customers with limited or no formal credit histories, thereby advancing financial inclusion while enhancing the accuracy, speed, and transparency of credit decision-making processes in digital banking ecosystems
 
Conclusions and Policy Implications: The findings of this study underscore the transformative potential of integrating traditional and alternative data sources in retail credit scoring, particularly within the digital infrastructure of neobanks. Unlike conventional banks, which depend heavily on historical credit data, neobanks can leverage real-time transactional behavior, digital footprints, and behavioral analytics to evaluate customer creditworthiness with greater accuracy and inclusivity. The prioritization of factors such as financial transactions, legal and insurance records, and online activity suggests that dynamic, behavior-based indicators outperform static demographic variables in predicting credit risk.
Moreover, the use of the Best-Worst Method (BWM) ensures the structured, expert-driven weighting of indicators, enhancing model robustness and decision-making precision. Importantly, the model addresses the credit assessment challenges faced by underserved populations, particularly those without formal credit histories, by utilizing accessible, digital-first metrics.
In conclusion, this study contributes both theoretically and practically to the development of next-generation credit scoring models in fintech. For neobanks, adopting such hybrid frameworks offers a pathway toward data-driven lending strategies that are faster, fairer, and better aligned with the needs of digitally engaged consumers. Future research should explore the application of machine learning to further refine predictive accuracy and incorporate evolving forms of behavioral and contextual data.