Constructing Financial Trust in AI-Enabled Digital Credit Information Systems: A Grounded Theory of SME Adoption Under Algorithmic Opacity

Abstract

AI-based digitall credit information systems made possible by AI offer quicker and expanded finance to small and medium enterprises (SMEs), but the rationale behind their scoring is typically unknown to the applicants. This study posits the question of how financial trust is created when SMEs are required to approve consequential credit decisions that they cannot view the algorithm behind the decision. A secondary qualitative synthesis was designed based on grounded theory utilising a purposely segregated evidence structure. The literature review and methodology were sensitised using 20 studies published between 2016 and 2023, and a corpus of 15 different primary empirical studies between 2021 and 2024 was formed. Constant comparison, focused coding, memo writing, and theoretical integration yielded three themes: evidentiary legibility, procedural agency and contestability, and ecosystem-embedded assurance. Bounded authorisation under managed opacity is the core category that describes adoption as a temporary choice to trust a digital lender when controlled evidence limits the observance of a lack of transparency by knowledgeable evidence and conditions of practical appeal as well as responsible institutional frameworks. The analysis questions the premise that more technical disclosure is necessarily a generation of trust. Foodless endowments may form a misleading promise of security, and cursory consent may hide stigmatizing information procedures or feeble lines of remediation. Trust is thus tuned based on the system performance, control by the applicant, lender behavior and the business value realized. The research adds an information-systems explanation of SME digital credit adoption, which connects explicable artificial intelligence, financial technology, and governance. Here it also outlines design and policy priorities; reason codes associated with corrective action, human review of disputed decisions, data provenance controls, similar pricing, and on-going accountability of post-loan.