RSR Research: Retail Connections - Predictive Analytics in a Down Economy
Through a special arrangement, presented here for discussion is a summary of an article from Retail Paradox, Retail Systems Research's weekly analysis on emerging issues facing retailers.
At an analytics roundtable at RetailConnections' Third Annual Business Executive Summit, I had a really interesting exchange with the CIO of a major jewelry retailer. I asked, "What good could predictive analytics possibly have done for you in 2009? After all, no one had a clue what demand was going to be."
The answer I got was fascinating. In this CIO's view, predictive analytics had been completely helpful to his company: sales were trending low, so retailers bought based on those trends, and essentially sold out to the walls. In many ways, 2009 was a very profitable year, with little excess inventory.
That really gave me pause. All year we thought retailers were under-buying. And no one really predicted aggregate holiday sales with any degree of accuracy. The NRF thought "flat"; PwC: "up one percent"; me: "up two percent." I think the final number was, in fact, up two percent, but not driven by the categories I thought they'd be. So if we couldn't predict aggregate demand with any sense of confidence, how could an analytics engine predict sales by SKU?
I suppose the difference is in the desired sell-through rates. Average seasonal sell-through rates are typically about 65-75 percent. Apparently retailers gave themselves the opportunity to raise that rate into the 90's. And it was good. Yet this year, port traffic is back up, and retailers appear to be buying up again. Even so, no one seems to have their arms around aggregate demand yet. So have we decided to drop sell-through rates again? If so, why? If you have some thoughts about this, please do let me know. It's one of those retail paradoxes that I can't quite figure out.
Discussion Questions: What do you think of the value of predictive analytics? How does such analytics overcome uncertainty in demand?