Gregory Piatetsky-Shapiro, Sep 27, 2010
In 1996 Tim moved to Dow's Marketing and Sales Expertise Center where he led the development of Dow's Customer Loyalty program and worked on Marketing Research. In 2005 Tim became the manager of Dow's Advanced Analytics group. Tim has over 100 internal papers and published 15 external papers. He has delivered numerous talks at various quantitative methods forums. Recently, Tim co-chaired SAS F2010, Business Forecasting Conference.
Tim's group is among the first users of new business-user-oriented SAS® Rapid Predictive Modeler (SAS RPM), and I spoke with him about SAS RPM and his analytics work.
Gregory Piatetsky-Shapiro: Can you describe in a couple of paragraphs what is your job at Dow ?
I am the Director of a group called Advanced Analytics. It is our responsibility to aid businesses and functions in decision making with the use of mathematics. We like to think of it as a "rigorous second opinion". We use various mathematical, statistical, machine learning and operations research based approaches to predict, forecast, simulate or optimize a decision support problem the business or function might have. Personally I promote the use of Advanced Analytics;
I help ensure we have people, technology (hardware and software) as well as data resources as infrastructure to help solve these problems; I find the projects; I assess methodologies; and, just for fun, I do modeling as well.
GPS: What is the role of Advanced Analytics at Dow ? In which areas of the company are they used? What are the most important analytic questions?
Tim Rey: As with any other resource, we are here to get more value out of the data investments we have. We have a very large IT department that manages our data infrastructure. So our role is to help the businesses and functions leverage this internal along with various external data sources for corporate gain. It is all about the money. We are here to help increase demand, revenue and share or decrease cost. Dow has many modeling groups in R&D, Manufacturing and Engineering already. Couple that with over 9000 JMP users, we have a lot of people applying quantitative methods. Our job is to work in the "higher end" of the risk, uncertainty and complexity spectrum. We have solved problems in Strategy, Finance, Auditing, Supply Chain, Purchasing, HR, Growth as well as in many specific business units.
Currently we have a huge influx of forecasting work....this takes the form of simply univariate forecast models, more complex exogenous models (ARIMAX) as well as full fledge econometric structural models or even system dynamics models.
We keep track of our past forecasts and how correct we were by comparing the model/hold out/out of sample model metric (normally SMAPE) to the rolling SMAPE for new forecasts.
GPS: What analytic tools are you (and your group) have been using in the past year? How long have you (and your group) been using SAS Rapid Predictive Modeler and for what tasks?
Tim Rey: We are a pretty heavy SAS shop. We use SAS Forecast Server, SAS Enterprize Miner, SAS Enterprize Guide, SAS OR and JMP along with Matlab, iThink, Crystal Ball, Risk Solver and a mess of our own Matlab code (Support Vector Machines and Genetic Programming per se).
We were an early tester of SAS RPM, and are just starting to use it more.
GPS: How does using SAS RPM compare with SAS base and SAS Enterprise miner ?
Tim Rey: SAS RPM does a couple of things. First, it provides more Data Mining process structure than something like JMP. Second, it makes it easier for a beginner to get "something"....It takes you further in true Data Mining than a BASE SAS or SAS EG can. It does not replace the utility of a full SAS EM. It also uses the traditional Train, Test, and Validate split, so it helps users to avoid the mistake of testing on train data.
I expect that SAS RPM will increase the demand on our analytics group, but it is too soon to tell.
We expect increased use from the 250 Data Miners we have trained, from the 7500 JMP users or from the 25 core data miners/modelers.
GPS: What advice would you give students who consider career in analytics/data mining?
My daughter is a undergraduate in Math (smile)...... so I am keen on this....
1. Be patient and be willing to get an advanced degree (your competition is and does!).
2. It is a growing and exciting market with lots of opportunities.
3. Look for internships during the summer and don't be afraid to travel to a new and exciting place.
4. Diversify your studies (include Statistics (Prediction and Forecasting), Machine Learning, OR (optimization and simulation), pure Math, statistical graphics, etc.
5. Learn how to work with data (lots of it) (SAS, Sequel, etc.)
6. Sharpen you communications skills, this is a consulting job!
GPS: What is a recent book you read and liked?
Tim Rey: Ahead of the Curve by Joe Ellis (A Commonsense Guide to Forecasting Business and Market Cycles)
GPS: What do you like to do outside of work?
Tim Rey: I enjoy spending time with my family and I am a big fisherman and hunter.