Beyond timeseries
When all you have is a hammer, everything looks like a nail. The hammer long favored by the supply chain community has been timeseries and, as a result, in supply chain circles all problems look like timeseries forecasts. The hammering temptation is compounded by the extensive literature that exists on timeseries forecasting beyond supply chain use cases. It’s not just the hammer that we already have: we are also in the middle of an entire shopping mall full of shiny hammers, featuring all shapes, sizes and colors.
However, timeseries are largely inadequate to reasonably model anything except, maybe, the simplest supply chain situations. As a result, the best timeseries forecasts, no matter how accurate, are routinely defeated by the mundane aspects of supply chains. Yet, when facing these situations, the community’s instinctive reaction the companies operating supply chains, their software vendors, and professors teaching supply chain  is to look for more accurate forecasts. After all, what else could there ever be but more accurate forecasts?
The most difficult step to mentally move outside the timeseries box is to acknowledge the very existence of the problem  i.e. the limitation of the timeseries themselves  without (yet) being able to exhibit an alternative solution. Indeed, the history of science tells us that problems tend to be “unthinkable” until a solution is found. Problems devoid of solutions^{1} are usually dismissed as irrelevant. Unfortunately, unless we start by positing that a solution might exist, we can’t even start looking for one.
Let’s have a closer look at the timeseries perspective and its supply chain use cases. So far, I have introduced three supply chain personae respectively named Paris, Miami and Amsterdam (more will follow). These personae represent fictitious, yet realistic, depictions of realworld supply chains. Can timeseries be used to reflect something vaguely approximating the “demand” in either of these three situations? For each one of these three situations, the answer is negative:
 Paris, a fashion network, involves massive substitution and cannibalization effects. The essence of the mechanism at play, the fuzzy perception by the clients of the assortment as a whole, is lost when timeseries are adopted.
 Miami, an aviation MRO, involves AOG (aircraft on ground) incidents, where a missing part ends up grounding the whole aircraft. Both AOGs and the cyclical nature of the part repairs are also lost when a timeseries forecast is adopted.
 Amsterdam, a cheese brand, is rigidly constrained on both the supply side and the demand side. As a result, while both supply and demand could possibly be represented as timeseries, the only pieces of interest happen in between those series.
Yet, supply chain textbooks are full of “examples” that involve timeseries analysis and timeseries forecasts. However, the validity of these examples is cause for concern. These examples feature nondescript companies that happen to produce and/or sell “something”, with zero specifics given. Yet, the devil is in the details. Whenever we start uncovering the fine print, as it is done in the supply chain personae introduced above, it becomes apparent that the timeseries perspective is essentially a collection of toy problems, which are going to keep students and professors busy, but aren’t really fit for any realworld use.
The timeseries perspective is one of the root causes^{2} that explains the ubiquitous use of spreadsheets in supply chains despite the availability of Advanced Planning Systems (APS) for three decades in most large companies. Supply chain practitioners are reverting to their spreadsheets because the APS is failing them.
The specific case of forecasting accuracy is interesting. Practitioners aren’t capable of beating APS accuracywise (expect maybe the truly dysfunctional ones). This has been the case for decades. Even in the 1990s, reasonably tuned parametric timeseries models were already beating humans accuracywise. The reluctance of supply chain practitioners to give up on their spreadsheets cannot be explained away by their reluctance to change, not over three decades.
A fundamental design issue is in the APS themselves, such as gearing the entire piece of software around time series, which misfits the problem  offers a vastly simpler explanation and more compelling explanation. However, this leaves us with the problem of why so many companies did adopt APS (frequently, several ones) if APS delivers so little.
This counterintuitive situation is a case of street light effect.
Timeseries are failing supply chains and yet, as it’s difficult to think of anything else, practitioners, and their management, frequently hold the default opinion that the timeseries perspective is what they actually need; despite their daily routine and the heuristics implemented in their spreadsheets contradicting this opinion. Moreover, data visualization concerns tend to be conflated with data modelization concerns. No matter which modelization perspective gets adopted, timeseries are a visualization mechanism that is likely to remain, indeed human vision is mostly 2D, and that most supply chain matters involve time as a dimension of interest. It’s not because an approach is good for visualization that its benefits will automatically apply to modelization.
The purest supply chain form of the timeseries paradigm is probably Flowcasting which reifies the entire supply chain as a collection of timeseries. Based on discussions with supply chain directors, it appears that flowcasting has failed every single time it was attempted. Putting the timeseries front and center was clearly an aggravating factor.
So far, in this post, no alternative to timeseries and timeseries forecasting has been proposed. Yet, this is the crux of the streetlight effect: once you know that you’re not looking in the right place, you should be looking elsewhere, no matter how dark those other places might be.
In my series of supply chain lectures, I will be gradually introducing elements to move beyond the timeseries paradigm. These elements reflect directions that Lokad started to take years ago. However, I invite my readers to try to imagine what their supply chain practice could look like if they were operating beyond the limitations of the timeseries paradigm.

In the early 1990s, mail order companies had all the supply chain infrastructure in place to become ecommerce giants. Yet, the mail order catalogue  the solution to creating offering awareness among consumers  had been around for so long that those actors had nearly all lost sight of the problem they were trying to solve: to sell at a distance. New ecommerce entrants became market leaders while they initially had very little competitive edge, especially as far as their supply chain infrastructures were concerned. ↩︎

One other major root cause, beside timeseries, is the deterministic perspective adopted by APS. The future is assumed to be perfectly known, leaving no place for uncertainty. However, uncertainty is irreducible and needs to be frontally addressed. Lokad does this through probabilistic forecasting, however this concern is largely orthogonal to the timeseries concern. ↩︎