Uncovering previously unknown information in business data is a core element of business intelligence applications. For instance, modelling banking transactions is widely used to not only enhance business performance but also as a security measure. Quite often potential information arises in the form of data clusters. However, challenges and opportunities appear to home in onto the determination of exactly which clusters contain 'interesting' information. Cluster optimisation still remains a major challenge within the data science community. We propose a two-phase algorithm that starts with cluster identification and optimises clusters by testing cluster parameters. To attain optimisation, dimensional reduction methods are carried out iteratively, measuring and testing each pattern for significance. Using transactional banking data, the algorithm sets cluster optimisation as a basis for identifying banking transactional behaviour. The novel method presents potential extensions into forensic investigations in fields such as accounting, criminology, engineering and many others.