A Forecast for Change

Keith Collins
CTO, SAS

For as high as it ranks it in magic quadrants, business intelligence (BI) market share and flat-out leadership in advanced analytics, SAS has never been a company to dwell internally on titles or pecking orders. So it was only a little surprising when we first met SAS’s elusive CTO Keith Collins and asked when he’d earned his senior title that he replied, “Um, I don’t recall exactly, can we can get back to you?” A natural fit for SAS’s homegrown culture, Collins joined the company in 1984 as a tech liaison just two years out of North Carolina State University. Rising through R&D, Collins eventually led the effort to the flagship SAS 9 platform and founded the company’s CIO Consumer Advisory Board in 2004. Only this year has he been able to return to his R&D roots where he is once again “living and breathing BI.” A busy man with a strong outside interest in education (he sits on university and museum boards), Collins still found time to share a thought or two with DM Review Editorial Director Jim Ericson.

DMR: SAS seems to be doing a lot of new things with bundled analytics.

Keith Collins: Let me back up for a definition so you know where we are coming from. When we use the word ‘analytics,’ that’s a bucket for us that includes areas of classic statistics, predictive modeling, forecasting and optimization. They’re all in the bucket I’m referring to.

DMR: Okay then, let’s dive into an area like forecasting where there are some new products.

KC: The whole area of forecasting is changing, for example in the retail industry, where they are ramping up technology beyond the use of POS [point-of-sale] terminals. They’re beginning to focus on how to optimize their businesses with better-quality forecasting that deals with seasonality or requires an order of magnitude more computational power. At the same time, a lot is happening in areas of optimizing the processes and the flows. One of the things you’ve probably seen from us is this shift to actually provide not just tools but also packaged analytic solutions, and our move specifically into retail is a great example of that.

DMR: We know all forecasts are wrong and tend to be slow, labor-intensive and manual, so what is the qualitative advance you are bringing?

KC: One shift is the high-performance forecasting we’ve brought to the market, which has changed the scale of what people can accomplish with the computing resources they have available. They can forecast more frequently across a larger range of products. The other shift is putting an appropriate interface around forecasting and moving away from, ‘I have to understand and write for an ARIMA [AutoRegressive Integrated Moving Average] model, applied to time series data’ to ‘We now have enough resources to generate multiple forecasts and deliver the best one to the user.’ We’ve shifted the paradigm away from the highly skilled technical capability required to create a forecast. Now we’re looking at the data, generating multiple forecasts, picking the best one and deciding whether to go with it or override it. The process of understanding the override actually helps you drive the constraints for subsequent forecasting.

DMR: When you say multiple forecasts, are you referring to scenario testing?

KC: Actually, I’m referring to the high-performance engine behind the forecast server, behind the user interface people interact with. Depending on how frequently you get information or at what level of the hierarchy it resides in, you may have a lot of information about something and very little information about something else. The product runs through different analytical methods and picks the one that’s most appropriate for that particular SKU or item at any level of the hierarchy. The high performance comes from the fact that it can automatically generate many different forecasts using different analytical methods and pick the best one, depending on the metrics you choose.

DMR: You’ve also released a new data visualization package.

KC: Right. The core of the desktop framework is a product called JMP that is very focused on visualization and the application of analytic methods to data. It’s a very powerful tool that’s tied together with the SAS platform to access the data, the data integration and data quality pieces. It harnesses the data manipulation capabilities and analytical computing power we’re talking about for predictive analytics, statistical modeling, optimization and the rest. The connection between JMP and SAS lets you marry the best in desktop visualization with enterprise class data integration and business analytics.

DMR: Speaking of computing power, we’re hearing about companies starting to offload noncore infrastructure and applications.

KC: It is already changing. We have a SAS OnDemand offering, a business unit that manages that process, and it is growing year over year without us even pushing it as an objective. We call it ‘SAS OnDemand’ because we don’t want to get into debates about multitenant services, but it’s really driven by how quickly can we help a customer address time to market or skills and capability needs. We host it on our premises and, at any point in time, they can choose to bring it in house.

Page 1 of 2.