DALE HAGEMEYER: Let’s start by defining predictive modeling as practiced in trade promotion management today. By way of analogy, consider: Most cars now have a little gauge that tells you how many miles you can drive before you run out of gas.
Just how does it know? Well, unbeknownst to you, there are algorithms running inside. These systems “know” how much gas your car uses, depending on how it’s being driven. It knows, for example, that if you’re towing a boat or driving aggressively, your fuel consumption will go up, and therefore you will have to get gas sooner. We might call this model “PHD” — “Push Here, Dummy.”
By contrast, in the old days before these systems came along, you could have had someone –– a Ph.D in statistics, say –– ride along with you in the passenger seat and, every once in a while, get out with a dipstick and insert it into your gas tank. Then, using a slide rule and other mysterious tools of the trade –– and for a fee, of course – this expert would be able tell you, “You have ‘X’ more miles before you have to stop for gas.”
Now, in point of fact, trade promotion optimization is not a strict choice between “Push Here, Dummy” or a Ph.D with a dipstick.
Rather, it’s a way to simplify your management challenges by using predictive models and scenarios to help every person on your sales team function as if he or she were one of your best salespersons or team members.
In short, the value of predictive models is to reduce complexity, and to manage performance upward by giving your people some better tools.
Now, it’s certainly true that if your movement data at shelf point is thin, if you have products that aren’t moving, your predictability will be compromised. So we’re assuming robust data sets, which enable you to be more reliable.
It’s also true that no predictive tool can ever say: “This is the absolute one right answer.” Predictive models are analogous to a meteorological forecast that says, for example: “There’s a 60-percent chance of rain tomorrow.” It’s the person receiving that information, not the system, who needs to decide: “With a 60-percent chance of rain, am I going to the picnic or the game without an umbrella?”
The forecast itself will never be so binary as to say: “Take an umbrella” or “Don’t take an umbrella.”
To return to the gas-gauge analogy, think about the 16-year- old new driver looking at a gauge reading one-eighth of a tank. What does that mean to him or her? Probably not much. It’s meaningful only to those of us who have had the experience of continuing to drive for too long, running out of gas, and having to call Mom or Dad from the side of the road. A person with more driving experience will look at the gauge and say: “An eighth of a tank means I can go 20 miles; I know I can get where I’m going.”
ALEX RING: I will take three shots at this point of view. I agree that modeling and optimization depend on data that is “robust.” Now let’s consider “reliable.”
Take that event you mentioned. Are you really going to lock in your plan to leave your umbrella at home based on a weather forecast seven days out? My guess is no. The forecast is simply not that reliable.
References:
Archives