“Did you hear about the 6 foot tall man who tried to walk across a lake with an average depth of 4 feet? He drowned." - Author Unknown

Jakob Bernoulli (1645-1705) was a Swiss mathematician and is credited with Bernoulli’s Law, known more colloquially as the “Law of Large Numbers” (not to be confused with Daniel Bernoulli’s Law from the realm of physics, which has to do with lift and fluid dynamics).

The law simply states that the more times you repeat a random event, such as flipping a coin or rolling a die, the outcomes will trend closer and closer to the theoretical average. Bernoulli described his law as being “so simple that even the stupidest man instinctively knows it is true.”

We’ve heard the same argument at least a dozen times since Flowcasting the Retail Supply Chain was released last year: Why would you even suggest forecasting weekly item demand at store level? Aren’t aggregate forecasts of larger numbers more accurate?

This viewpoint is best summed up in an online e-book (provocatively titled “What Jessica Simpson Has to Teach Supply Chain Executives”) that is semi-critical of Flowcasting:

“While Flowcasting validates the importance of the store, having a forecast model at the store level is ridiculous. This is where a forecast is the least accurate. Can you imagine the devastation to a business if it tried to forecast the purchase of each SKU at a store level? Basic statistics show that forecasts are most accurate at a consolidated level, such as a country or distribution center, and even still cause inordinate levels of shortages and excess inventory.”

In other words, why is it that we haven’t grasped “what the stupidest man instinctively knows is true”?

This line of reasoning betrays a fundamental and critical flaw in thinking when it comes to forecasting in the retail supply chain.

We’re not saying that forecasts of aggregate demand aren’t more accurate than the far more granular forecasts we advocate in Flowcasting – they most assuredly are. We’re just saying that this fact is completely irrelevant.

Retail out-of-stocks have been stubbornly stuck at 8% since they started measuring it (15% when an item is promoted). The methods and samples vary, but all of the major studies* point to store level forecasting and replenishment practices as the primary cause for the out-of-stocks.

So, given the fact that: 1) consumer demand occurs by item by day at store level; 2) retail out-of-stocks occur by item by day at store level and; 3) the root cause of the out-of-stocks has been shown to be at store level, what’s the purpose of producing “more accurate” demand forecasts at aggregate levels?

If you take the aggregation argument to its logical conclusion, you would just forecast annual GDP for your country as a whole. When you aggregate that much, a ton of random variation would be smoothed out and you’d surely get error rates in fractions of a percent. But how would you use it to run a retail business where consumers are buying particular items in particular stores on particular days?

In our opinion, the lack of effort and thought about store level forecasting and replenishment practices over the last decade – despite strong evidence that this is where the opportunity lies (everyone seemed to be hiding behind the Law of Large Numbers) - is the main reason why out-of-stocks haven’t budged.

A view from 10,000 feet may be beautiful and majestic, but it doesn’t help you when you’re looking for a needle in a haystack.

* “Where to Look for Incremental Sales Gains – The Retail Problem of Out-of-Stock Merchandise”, a 1996 study conducted by Andersen Consulting for the Coca-Cola Retailing Research Council, “Retail Out-of-Stocks: A Worldwide Examination of Extent, Causes and Consumer Responses” conducted by the Food Marketing Institute and Grocery Manufacturers of America in 2002 and most recently “Improve OOS Methods at the Shelf”, a 2006 study by Tom Gruen from University of Colorado and Daniel Corsten from London Business School.