How does a $55 billion company get a sense of its demand when it operates in virtually every market – from paint to electronics – on a global scale?
That’s the question Timothy D. Rey answered during a presentation on how his company, Dow Chemical, used a data mining approach in order to answer the forecasting problem of how to sense demand in the marketplace. “When you’re that complex, in that many markets and in that many regions, you end up having to use a lot of input variables,” Rey said.
As it turned out, Dow whittled down more than 15,000 variables to create 7,000 models in a six-month time frame in order to forecast in areas like volume, inventory and net sales. In particular, Dow used external time series data, as opposed to internal data to look ahead, with hopes of “using a forecasting model in a whole mess of different areas at the company.”
For an in-depth understanding of how Dow did just that, check out the archive video of Rey’s presentation below.
Highlights from the presentation include:
- Time series data is easy to obtain, although it may mean paying for it.
- Organizations can’t just look in “the rear-view mirror at your own internal history,” said Rey. “I think in business we can be very shortsighted and let our eyes deceive us when we don’t use external data, when we don’t use the future and we stick with just using structured, not unstructured data.”
- The “75 percent” problem: data prep time is consuming and complex, often taking up 60 to 80 percent of a data mining effort. But then “the fun stuff starts,” Rey said of building models.
- Like SAS’ Analytics Lifecycle, which identifies different people’s roles during the data mining process, Dow adopted its own, including data extractors, data harmonizers and advanced analytics modelers. In the end, Dow was able to reduce costs, be more agile in the market and improve the accuracy of its models.
You can also read more about how Dow Chemical uses SAS.