Business Intelligence SIG: Mining huge customer datasets to understand cause and effect



  • Enterprises receive a flood of data from their customers: purchases, web browsing, calls to call centers, mobile data, and so on. The customer is clearly signalling their needs, intentions and motivations, but the data is too vast and messy to analyze. When this data is summarized in a warehouse the precious causal relationships - what drives customer behavior - are lost. The real value is in the detail, which means keeping and analyzing all the data.

    In this Big Data era it becomes possible to store all this granular data and analyze it efficiently to discover cause and effect. Causata, a Bay Area startup, is attacking this problem. This requires organising many terabytes of data for millions of customers, with distributed analysis and machine learning executed in parallel close to the data. Causata structures all the granular multichannel customer data in a single timeline for each customer. The timeline is dynamically queried to find these points of interest, and predictive records are constructed which summarize the customer's behavior leading up to, and following, that instant. Statistical learning over these predictive records tells us what is driving customer behavior.

    Of course correlation is not causation. Analyzing customer data tick-by-tick generates powerful hypotheses of what drives behavior. This learning then drives automated real-time decisions on how to interact with a customer - what to show them on a website, what discount to offer them, whether to accept an out of warranty return, what products are they likely to buy next. Causality can be studied by introducing experimentation into these real-time decisions - allowing the learning of true cause and effect - and feeding this learning back into making better decisions.


    6:30pm - 7:00pm - Registration & Networking

    7:00pm - 8:30pm - Presentation