The wrong lessons of a supply chain crisis
Whenever humans are involved, complex systems are never amenable to “easy” models, if only due to second order effects 1. Supply chains are no exception, and yet these models are prevalent 2, leading to all sorts of inefficiencies such as inventory write-offs, waste, delays, unused capacity, poor customer service, etc. Yet, these inefficiencies are not the biggest problem associated with simplistic models, the biggest problem is that they prevent people in charge from even considering what could eventually become proper remedies. Indeed, if a problem is already solved, why would it need to be solved twice?
Approaching this subject in supply chain is difficult because every supply chain has its own unique combination of clients, suppliers, processes, machines, software, people, regulations, etc. As a result, it usually takes a lot of time, weeks if not months, sometimes years, to start grasping what makes a specific supply chain tick. Furthermore, key information about the supply chain is usually considered as a trade secret. This makes it very hard to have an intelligible public communication about any given supply chain, as it’s such a haphazard collection of obscure details - notwithstanding those that can’t be communicated at all.
However, beyond the tragedy, the SARS-CoV-2 pandemic offered quite a spectacular illustration of all the things that tend to go wrong when pseudoscience meets power structures. Unlike most privately dysfunctional supply chains, the events afferent to this pandemic have been widely publicized, and interestingly enough, many of those elements were supply chain problems. Also, as states were involved - arguably the biggest organizations of all -, the problems ended up being magnified.
Let’s review some of the most salient dysfunctions 3 of the response to this pandemic, establishing the relevant parallels with supply chains.
Trust the evidence.: Until early March, the dominant position of many Western governments (including the USA and France) in regards to the pandemic was a complete dismissal of the problem. It was a Chinese problem after all. Within a few weeks, the same governments - and to be fair, most of the political class - seemingly discovered the very notion of exponential growth, and based on those models, went from “nothing to see” to supporting the most intrusive policies ever implemented by any Western government since WWII. Many governments (the USA and France included again), assisted by their administrations, went out of their way to derail many (most?) pandemic-related initiatives, first when it came to assessing the progress of the pandemic 4 and second when it came to assessing the effectiveness of specific treatments 5. The road to hell is paved with good intentions.
Most of those measurement woes were the direct consequence of decisions and processes supposedly grounded in science 6. However, complex systems can never be resumed by a handful of variables. As a result, measurements (i.e. gathering evidence), requires both practice and culture. Pouring numbers into spreadsheets or a similar piece of software is only a tiny fraction of the story. The practice is difficult, and it typically takes months - if not years - for the practice to mature so that it can earn any degree of trust. Also, numbers only make sense in the mind of the observers. No matter how “neutral” the evidence might be, observers are, by definition, subjective.
Yet, when facing a crisis, it’s all too tempting to just draw conclusions based on “evidence”, while the so-called-evidence is numbers that have been hastily gathered without a deep culture of proper measurement in place. Leveraging hastily collected numbers isn’t being “rational” or “scientific”, but it pretends to be, which makes the whole thing intellectually dangerous.
Too many supply chain executives taking over a new division start with a quick assessment of their “service levels” to immediately jump to the conclusion that whatever this number is, it should be brought closer to 100%, disregarding whether this undertaking actually makes sense for the company. For example, achieving near perfect service levels on strawberries is certainly possible if one company is willing to systematically waste half of the production, due to the short shelf-life of the fruit.
Similarly, too many companies engage in multi-year upgrade plans with their software vendors, which is nonsense timewise: by the time the software gets deployed, the product is already outdated. Not only the measurements must be methodologically correct, but they should be compatible with an action plan that happens fast enough for the whole thing to remain relevant.
The solution lies in creating a culture of relevant measurements: measurements that are apt to guide the immediate action, or at the very least, apt to provide fresh insights. Indeed, when facing a system as complex as a supply chain, it’s easy to get lost in a sea of KPIs. Conversely, it’s also easy to overemphasize a few indicators that don’t tell the whole picture. Sorting out what is relevant to measure and what is not, as well as turning measurements in course into actions, is almost entirely left to the subjectivity of the teams in charge. Yet, it’s the core of the scientific (rational) process, not the measurements themselves, which are merely a by-product of the undertaking.
Adapt the process. Face masks were originally deemed superfluous - after all, France kept destroying its own (freshly expired) stocks of medical masks until February - for them to later become mandatory in May. In between, mask shortages occurred, and obvious options such as safe reuse via decontamination 7 were mostly ignored. Overall, many bureaucracies went out of their way to make the situation worse 8. Besides France, multiple countries went down a similar path. Such a complete reversal of policy - in a mere matter of weeks - hints at a strongly irrational decision making process. Indeed, medical masks - there are various flavors - have been in use for decades. While science kept progressing during this pandemic, there was no notable breakthrough in regards to the understanding of the degree of protection (or lack thereof) offered by masks during a viral epidemic. Thus, whatever changes took place weren’t driven by science, no matter which option is the correct one (i.e. masks or no masks).
The adherence to a process or a norm is the default option, just because it would be too exhausting to challenge everything all the time. Usually, there is no grand strategy or plan involved. Governments, companies and even individuals are creatures of habit out of necessity. At an individual level, stability is needed for psychological safety. At a group level, stability is required for the group to function at all.
However, changing the process might unlock better, safer, more profitable ways to operate, and thus, it’s tempting to do so. After all, immobilism isn’t going to get anybody anywhere. However, when it comes to complex systems, it is important not to downplay the degree of unknowns that are invariably associated with the unintended consequences of a change of process. For example, it appears that both in the USA and France, having the combination of (a) the end of two months of hard lock-down, (b) entire populations of masked individuals walking the streets and (c) disturbing police brutalities did end up creating a context prone to riots. It’s safe to say that riots weren’t really intended by the respective governments on both sides of the Atlantic. They fall in the category of unintended consequences when dealing with a complex system - the society at large in this case.
The danger here is to overestimate how much we truly know about a complex system. Thus, when revising an existing process (i.e. “mask wearers being the exception not the norm”), it’s important to assess what are the benefits associated with this process, even if those aren’t obvious. Frequently, there is wisdom to be found in things that have been done in a certain way for a long time. Even dysfunctional processes tend to have a few qualities - otherwise they would never have managed to stick around in the first place.
Most companies don’t question their own supply chain processes enough, and as a result don’t really know either their strengths or their weaknesses 9. For example, most ERP migration projects are started on the premise that the “legacy” ERP is the problem. Sticking with a decade old system is akin to immobilism, thus, something has to be done. Years later, when the next ERP is finally deployed, the company realizes that this ERP does not deliver the expected benefits. Invariably the integrator, the ERP vendor, the internal IT teams, or the whole lot of them get blamed for the situation. Companies rarely (never?) acknowledge that the legacy system could have stayed in place just fine.
The reality is that improving processes with regards to a given strategy requires a fair bit of trial-and-error, precisely due to the relative impredicatibility of the supply chain - the complex system of interest. It frequently happens that an insight proves to be directionally correct (e.g. we have too many planners manually interfering with the forecasts all the time), while the “solution” turns out to be making the problem worse (e.g. let’s replace all those planners with this fantastic new A.I.). Tinkering with the process is acceptable, as long as it’s done in a capitalistic manner, i.e. as long as people recognize their errors and learn from them.
Take the safe option. Facing the risk of having hospitals overwhelmed due to the SARS-Cov-2 pandemic, the region of Paris called for a strict stay-at-home policy when facing all but the most dire symptoms, i.e. don’t seek any treatments unless it’s a life-and-death matter. In complete opposition, the region of Marseilles - the second largest urban region in France after Paris - maintained its usual policy to treat all patients who showed up at their hospitals. As the dust has basically settled down - in France at least - it turns out that the region of Marseilles has had roughly 5 times less pandemic-related deaths per capita than Paris. By adopting the “safe” option - again, based on seemingly hard scientific evidence - a region (Paris) did end-up making a bad situation vastly worse than a highly comparable region (Marseilles).
The fallacy at play here is the notion that “safe” options exist. They don’t. Life is risky, irremediably so. The “safe” option (stay-at-home) was in fact trading a specific risk (SARS-CoV-2) against many other non-quantified risks, including hard-to-treat complications due to delayed treatment of SARS-CoV-2 itself. The real world is all about balancing conflicting risks.
Interestingly, most people, at an individual level, are reasonably capable of doing a rough risk assessment for their own person, and are even willing to accept a certain degree of risks - such as driving, a well-known cause of serious, sometimes deadly, accidents. On the contrary, large organizations, and in particular their bureaucracies, struggle to deal with any degree of risk. The larger the organization, the more acute this problem is.
This leads to the seemingly puzzling behavior where large organizations tend to entirely dismiss emergent risks, until a specific point of time is reached where the risk cannot be ignored anymore - usually due to some external pressure, such as the press. At this point, the organization brutally adopts a new policy intended to eliminate the newly identified risk, frequently at the expense of all the other risks that did not get the same “press coverage”.
Supply chains usually entail some bureaucracy of their own, and their responses to risk are frequently similar: they ignore problems, potentially for years (such as neglecting transport optimization) to suddenly become overzealous about one narrow angle such as the CO2 footprint, while ignoring tons of other risks in the transition - the incorrect assessment of the CO2 footprint being one of them.
Part of the solution lies in embracing risk both conceptually and quantitatively. Conceptually, being “at risk” should be acceptable, i.e. it should stop being a taboo. Indeed, due to real world risks, a supply chain is always “at risk” somehow. It’s not because the risk is not identified or taboo that the risk ceases to exist. Quantitatively, risks should be roughly quantified. The “rough” part is important, because overstating / overplaying the knowledge about any given risk carries the risk of ignoring other risks.
In conclusion, some of the mismanagement of the pandemic that we have seen play out on a grand scale had little to do with the individuals that happened to be elected at this point in time, and was a lot more about modern delusions such as (a) when quantitative models get applied, it’s automatically “scientific” and (b) risks can be eliminated through direct cause-consequence course of an action. Mitigating those delusions in the future will be more a cultural process than replacing a few high-profile individuals from their positions.
In this grand scheme of things, events unfolded in ways that are eerily similar to what happens too frequently to supply chains when attempts are made at fixing problems while facing a crisis. Authorities emerge to both warn of (some of) the risks and to show the “safe” path. Simplistic models are adopted based on hastily gathered numbers. All of this validates the vision of the authority. Massive process changes are rolled-out in a hurry, but the most important pieces only arrive after the battle. Little attention is paid to the details, and tons of things fall apart. People find their way by breaking some of the processes, and a new normality emerges, only a bit more complex than the previous one due to the extra rules that have been adopted.
More than ever, the solution lies in a culture of healthy skepticism, cultivated common sense, and humility with regards to complex systems that don’t lend themselves to simplistic uniform corrective measures.
If a retailer puts a product on promotion, the first order effect is a lower margin that will hopefully be partially recovered through a higher sales volume. However, as a second order effect, the promotion creates an expectation among clients who will try to buy “at a discount” in the future. People don’t just react to change, they adapt. ↩︎
ABC analysis, min/max inventory, safety stocks, decoupling points, open to buy, classic time-series forecasts, etc. All these simplistic models are mostly broken-by-design, and yet remain very popular in supply chain literature. ↩︎
Living in Paris, my perspective is largely skewed by French events. Also, as a long time follower of US events, a country where I have lived on two occasions in my life, it strongly influences my perspective as well. ↩︎
Both the USA and France turned out to be laggards - compared to many other countries such as Germany or Iceland - in their capacity to test their population for the prevalence of SARS-CoV-2. In both situations, prevalence tests were largely hindered by administrative complications. ↩︎
For instance, opting for randomized controlled trials lasting longer than the majority of past epidemics similar to SARS-CoV-2 (i.e. a few months) is an obvious nonstarter. By construction, the results, if any, will only become available after the end of the epidemic, which defeats the purpose of treating patients and could even be construed as quite unethical. ↩︎
France did completely dismiss all its veterinary infrastructure, which was perfectly capable of performing large quantities SARS-CoV-2 tests - literally using the same devices and reagents that the ones used for human tests anyway - as it wasn’t deemed “safe”. ↩︎
In the USA, it took weeks for the FDA to approve the “Duke Protocol” for safe N95 masks reuse after H2O2 vapor decontamination. In France, no reuse protocol ever got approved before the end of the pandemic. Instead, the USA - and France even more so - traded the quantified risk of masks reuse (measured as very low) against the unquantified but severe risk of N95 counterfeits, which became ubiquitous for a few weeks while the world was facing widespread N95 mask shortages. ↩︎
The price gouging laws hastily acted on hydroalcoholic solutions in many places, including states in the USA and in France, delayed the most obvious supply chain responses to the problem: let producers raise their prices. Some of those producers would have engaged in rapid production expansion, leveraging the higher prices to fund their capacity investments. ↩︎
Beware of neutral third parties, for these are rarely neutral. Many market analyst companies draw the bulk of their revenue from software vendors, and thus, are prone to exaggerate the vendor’s benefits. Many consultants underestimate the risks associated to change because they are primarily needed when big changes are rolled out, etc. ↩︎