Providing financial service to Micro, Small and Medium Enterprises (MSMEs) in Indonesia presents a challenge for small rural banks such as People’ Credit Banks. These banks are required to infer risks about customers’ loan repayment from structured (quantitative, financial) and unstructured (qualitative, non-financial) type of credit information. In this study, the complex nature of credit related information is contextualised and represented in domain specific way using the eXtensible Markup Language (XML). An approach that enables the application of wider selections of data mining techniques on XML data is utilized. Experiments are performed using real world credit data obtained from an Indonesian bank. The results demonstrate the potential of the approach to generate reliable and valid patterns useful for evaluation of existing lending policy.