Disclosed herein is a method of classifying financial transactions by usage patterns of a user. The method includes analyzing metadata extracted from the information associated with financial transactions in accordance with at least one business rule. Where the analysis includes sequentially analyzing the metadata using a constant-time lookup data structure, a Radix tree, a Lucene tree and fuzzy logic methods, until a unique identifier is found that is associated with the metadata. The metadata with the unique identifier is then added to the constant-time lookup data structure to update the constant-time lookup data structure. The transaction data is then classified based on the unique identifier.