For a Data Scientist in Finance/Banking, 80~90% of their time is taken manipulating and processing data. Most of the modern analysis with machine learning is constrained due to traditional SQL capability. SQL is only good at flat data (table) but it fails to handle complicated transactional data (nested, hierarchical, time series etc.) There is big trend and strong needs that financial institutions need to deal with granular transactional data, not merely longer summarized financial data. The traditional analysis against the transactional data was focused more on identifying "trans-actor" versus "revolver" which still largely studied summarized customer abstraction. Abundant customer behavioral information was ignored. During our presentation we will present how to drive more value from transactional data (using credit cards as an example) at high speed using GPU's: a. Explore and exploit feature engineering to see how customer purchase transaction patterns can be identified in a meaningful way, (this way very hard in the past). b. Explore Trend and Evolvement i.e. how to identify the early yet critical trends in customer behavior using credit cards, and make sense of the evolvement over time. c. Synthesis and integration i.e. present the operational decision process end to end for these new pipelines and identify lessons learned.