The rapid growth of digital banking and neobanks has made customer experience a central source of competition in financial services. In addition to survey-based methods, app-store reviews provide a rich source of authentic customer voice; however, much of the prior literature has analyzed these data mainly at the levels of topic and sentiment. This study uses large language models to extract a structured, multilayer view of customer experience from reviews of Iranian banking and neobank applications. The data were collected from reviews of 18 apps on Cafe Bazaar. After cleaning, 16,145 unique reviews were segmented into 21,091 semantic units and analyzed. For each unit, the study extracted experience dimensions, feedback type, sentiment, issue, pain point, strength, need, expectation, and feature request, and then aggregated the results at both market and app levels. The findings show that credit and lending, app technical performance, login and access, and account opening were the most frequent experience dimensions. The main pain points were failure to receive verification codes, face-recognition errors, unresponsive support, app crashes, and failure to obtain credit. In contrast, fast card delivery, app speed, and interface simplicity emerged as the key strengths. Responsive support was by far the most frequent need, while increasing transfer limits was the leading feature request. Customer-voice profiles also differed descriptively across apps in both dominant topics and the mix of praise, problem reports, and complaints. The study suggests that large language models can move review analysis beyond topic and sentiment classification by transforming unstructured qualitative data into a multilayer structure useful for customer-experience analysis and managerial decision-making.