Attending the Predictive Analytics World (PAW) Conference is truly a rewarding experience. Compliments go to Eric Siegel and the conference organizers for assembling such an interesting cast of case studies and speakers.
Day 2 kicked off with a key note from Kim Larsen from Charles Schwab & Co. on net lift (or uplift or incremental) modeling using SAS. The latter topic is a perennial favorite among data mining practitioners. Net lift models are designed to target undecided or swing customers (similar to targeting swing voters in an election campaign) to get incremental lift in revenue from marketing campaigns, while not targeting customers who would have bought it whether or not they received a promotion offer. The key learning from Kim’s session is to use appropriate variable selection and transformation methods while preparing data to maximize returns from Net Lift Modeling.
Next-up were my colleagues, Manya Mayes and Fiona McNeill, who shared a cross-industry text analytics case study and best practices for classification, sentiment analysis and search techniques in the SAS lab session.
I also hosted Birds of Feather topics on in-database analytics and rapid predictive modeling during lunch hours on each day at the conference. Other SAS topics included were Twitter analysis, sentiment and social analysis, and time series data mining. These topics generated good discussions in a short group sessions around trends, adoption and customer interest/readiness. For the first time at PAW, we saw presence from a data warehouse appliance vendor, Netezza (also a SAS partner). Netezza presented on their analytic appliance and invited Tonya Balan from SAS to underscore our commitment for SAS In-Database and highlight SAS Scoring Accelerator for Netezza.
The afternoon sessions were packed with more customer case studies and vendor lab sessions. Of particular interest to me were Deutsche Postbank and PayPal. The Deutsche Postbank presentation by Franz Hofner focused on how model retrospection (which customer segments of a model under- or over-performed) is useful to gain insight about customer behavior but more importantly suggests ways to improve the models. The PayPal presentation by Piyanka Jain focused on how to optimize product up-sell path across multiple dimensions (e.g. not only on adoption rate but also on incremental profit).
In between sessions and labs, Tonya Balan and I had one-to-one meetings with customers. Discussion topics revolved around customer business issues, product capabilities, challenges with analytic data preparation, and how to measure success.
Overall I would vote that the Predictive Analytics World conference was hugely successful. We gained a lot from our interactions with attendees and vice-versa. It was great to see innovative data mining applications, modeling best practices and overall growing interest in text analytics and in-database analytics.
