|Wiley E. Coyote is a super-genius whose sole purpose in life is to catch, and eat, the Roadrunner. With all the available resources of Acme Corporation at his disposal, you’d think the Roadrunner was doomed.It never seems to work out that way, however. Wiley E. Coyote is the victim of over-complication. His booby-traps and schemes are just too sophisticated for his own good. It’s over-complication that consistently lands Wiley E at the bottom of the canyon.I think we can learn a lot from Wiley E., especially when you think about what’s happening with respect to forecasting for the retail/CPG supply chain.
For decades people have been forecasting what should have been calculated. And since you went down that path, very logically I might add, you’ve had to add more and more complexity to account for the cornucopia of programs, formats and challenges that marketers, merchants and salespeople have dreamed up.
The result? Sophisticated algorithms and approaches to forecasting something that should never have been allowed into the forecasting engine. I find it both funny and quite sad to hear self-proclaimed supply chain experts beating their collective chests and telling everyone to use “downstream” POS data in their mathematical forecast models.
How exactly, Wiley E? The answer you normally get is nice and reassuring…”by using advanced analytics”. This kinda reminds me of another great cartoon character, Foghorn Leghorn…”I say, boy, it’s all too complicated for ya. Let me splain it to ya”.
Perhaps you’d be better served if you read and understood Dr. Joesph Orlicky’s startling and profound statement…”never forecast what you can calculate”. Joe made this statement in the 1970’s and it’s truer today than it was back then.
When it comes to retail/CPG forecasting, most of the experts are terribly misguided. Not only are they forecasting what should be calculated, but they’re pre-occupied with the wrong end of the supply chain. Why aren’t they talking about the retail store (or web portal, or markets) and speaking a common language…consumer demand. Real demand.
Why wouldn’t you just forecast by item by store and calculate the rest? Not only is that now possible, it also provides the most reliable forecast for all supply chain partners since all the variables are factored in, where they occur.
Plus, as an added bonus, you’ll get unprecedented visibility into your supply chain. Not what you have in inventory now and what’s on order. That’s chickenfeed. We’re talking about projected sales, receipts, inventory, capacity needs and so on.
Yup, when you only forecast where it counts, you get so much more. It’s like that extra cherry on top of the ice cream sundae!
But, you ask, “wouldn’t you need to be pretty sophisticated to forecast consumer demand at store level”?
The answer is: not really. You will need to recognize a selling pattern, trend, seasonality and flag past events and abnormal conditions. The beauty is that you’d be forecasting consumer demand, which has much fewer constraints and conditions attached to it.
Since the forecast is at store level, you will also need a way to predict and manage slow selling items – since retail is dominated by them. And the forecast shouldn’t produce small decimals for these types of products. Rather they need to produce integer-like forecasts that can then be calculated into the demand plans for the supplier of the store.
In addition, you’ll need a simple and effective way to update the forecast, in aggregate and then for each store, for things like promotions and/or market intelligence.
What about promotions? Surely they’ll need advanced analytics?
Years ago we tested it. We developed some causal forecasting equations and also let teams of experienced merchants and supply chain planners develop forecasts for a bunch of upcoming promotions.
The teams, on average, developed more accurate forecasts than the algorithms. Not only better forecasts but obviously more accountability in the forecasting process too.
What we found was that it was just too difficult to find math models to reliably explain all the inter-relatedness of expected demand. Yet, somehow, the teams could use historical information and their own judgment about the future, collaborate, and predict the future with reasonable accuracy.
It’s ironic. The farther away from consumer demand you are trying to forecast, the more constraints and difficult it is to do.
You have a choice.
You can try to forecast your way to greatness, or you can calculate your way.