So Simchi-Levi recommended a two-level segmentation strategy (I have sometimes seen him use three or four), in which the 19 fast movers with low levels of variability would be built to stock, likely reducing production costs but also enabling the company to improve total competitiveness through faster total lead time.
Many of the rest in the Long Tail should continue to be built to order, but (importantly) a broad swath should be "candidates for elimination," Simchi-Levi said, as they are eating away at profitability.
Some companies don't really measure the profitability of individual products. A growing number of companies use a technique called Gross Margin Return on Inventory (GMROI) that basically calculates the return received from the investment in inventory at a SKU by SKU basis. The results, as I have seen first hand many times, are often eye opening. (More on this soon.)
But even GMROI isn't enough, Simchi-Levi says. The approach is a start, but doesn't nearly capture all the supply chain costs associated with managing the long tail, many of which are "hidden." When you add in total supply chain costs (inventory carrying cost, the costs associated with forecasting and managing these high variability products, engineering effort, etc.), a significant portion of the Long Tail will actually be negative margin products, Simchi-Levi says. Of course, marketing will usually make the argument that the products are needed for other reasons, and maybe they are, but the analysis of the cost must be comprehensive and clear to guide a more informed decision.
The takeaways: (1) dealing with the Long Tail effectively requires supply chain segmentation strategies; and (2) there are better ways to show the true financial impact by detailing the direct and hidden costs of Long Tail SKUs comprehensively.
Joe Shamir agrees with Simchi-Levi that the first place to start is determining whether long tail SKUs can be pruned, but believes in the end that for many companies most of the long tail will remain.
"David is right, companies need for sophisticated tools to better understand and manage the trade-offs" between the cost of carrying slow movers and the impact on sales, customers, and other reasons for keeping many Long Tail SKUs. But in the end, many/most companies will still carry lots of them. And "the only way to manage them effectively," Shamir told me, "is to use optimization."
Shamir adds that the growth in SKU counts has a companion trend, which is that demand is also getting "lumpier," or more inconsistent. He notes for example that for many slow movers, demand in any given period is likely to be zero, especially (and this is important) as the point of demand that is being forecast keeps moving down the supply chain. For instance, in the consumer goods sector that means we are now starting to forecast at the individual store level, not the retailer's upstream DC. Parallel examples are occurring in other sectors.
"So, with lumpy demand, it is really just as much about the frequency of orders as it is the quantity," Shamir says.
He makes a persuasive case that traditional forecast and inventory planning models that dominate the landscape today can't do the job very well in serving the intermittent demand in the Long Tail. Specifically, he says that traditional approaches, which I will characterize as first calculating the forecast with one of the algorithms of the many available, and then using the forecast error to calculate the safety stock that will deliver the desired service level.
But he says, one cannot "serve" the intermitted demand of the tail by producing accurate forecast. "It must be served with adequate inventory," he says. That traditional approach "works OK for fast moving SKUs," but Shamir argues that a fundamentally different approach is needed for slow movers.
That involves modeling demand and inventory stocking requirements using "stochastic" techniques. That may sound like an overly techie term, and I will dive into this subject in more detail soon, but just consider it this way - it's all about probabilities. The forecast and associated inventory requirements needed to meet target service levels shouldn't really be a single number (inaccurate as it usually is), but rather the volatile demand behavior should be modeled statistically, in order to simultaneously provide both the forecast and the information required to calculate the correct inventory to guarantee service.
Shamir says that this probabilistic approach is simply the right way not only to manage uncertainty generally, but especially to address the challenge of modeling both quantity and frequency of demand. He can show you lots of math and real-world examples that support his thesis.
Supply chain segmentation, hidden cost analysis, probabilistic optimization: some weapons to take on the Long Tail.