Supply Chain Resilience Reconsidered
In the last few years, “resilience” has become one of those words that executives feel obliged to use in every second sentence. After every disruption – a lockdown, a war, a blocked canal – the same question comes back: “How do we make our supply chain more resilient?” The conversations are sincere; the answers are often vague. “More visibility,” “more collaboration,” “more agility” – words that sound positive but explain very little about what we should actually do differently, every day, when we decide what to buy, make, move, and price.
In my book Introduction to Supply Chain, I proposed a view of supply chain as the disciplined management of decisions under uncertainty, with economics – not service levels or utilization – as the primary yardstick. That view has shaped how I think about resilience. In fact, it has led me to a definition that is much narrower than what you will typically find in textbooks, consulting brochures, or certification programs.
What I mean by “resilience”
When I speak of supply chain resilience, I have something very specific in mind.
A company – and its supply chain – is resilient if it can withstand an unplanned, systemic shock that threatens the flow of goods, and then restore that flow to its prior state.
There are two important qualifiers in this sentence.
The first is “unplanned.” Many unpleasant events are not shocks in this sense. A peak season surge, a promotion that works better than expected, the fact that lead times are noisy rather than constant – none of this is new or mysterious. It may be difficult to model accurately, but it does not sit outside the realm of reasonable anticipation. If you run out of stock every Christmas, this is not a resilience problem; it is a planning problem.
The second qualifier is “systemic.” A single store losing power, one truck breaking down, one supplier missing a shipment: those are local incidents. They can be irritating or even costly, but they do not threaten the continuity of the flow as a whole. A major port closing for months, a regulatory shock that suddenly makes a category unsellable, a war that halts trade routes across an entire region – here we are in systemic territory.
Resilience, in my vocabulary, is reserved for these rare, high-consequence events that both (a) could not reasonably have been planned in detail and (b) affect a substantial portion of the supply chain at once.
Everything else – the daily noise of demand, the sloppiness of lead times, the normal dance of promotions, the predictable antics of competitors – should be addressed through good supply chain practice, not framed as “resilience.”
How the mainstream talks about resilience
If you look at how large technology vendors, professional associations, and policy organizations describe resilience, you will see a different picture.
A fairly typical definition presents supply chain resilience as the ability to anticipate, adapt to, and recover from disruptions while keeping operations running. The emphasis is on continuity: the system should keep delivering acceptable service levels even when something unexpected happens.
The main professional bodies add another nuance: resilience is the capacity to return to a position of equilibrium after performance has deviated from expectations. In that framing, resilience is about coming back to “normal” after a disturbance, and it can be improved by having more response options and acting on them quickly.
From there, a familiar list of levers appears over and over again. Redundancy in the form of extra inventory, backup capacity, and alternative suppliers. Flexibility through multi-skilled labor and adaptable production. Visibility and collaboration, often enabled by digital platforms, to detect issues earlier and coordinate responses. Recent policy and consulting reports add another layer: the need to “balance” efficiency and resilience, sometimes by reconfiguring networks, adjusting sourcing footprints, or investing in new technologies. Academic reviews, particularly from the inventory-management perspective, catalog strategies such as stockpiling, multi-sourcing, capacity reservation, and flexible contracts under the heading of “resilience strategies.”
There is nothing absurd in this mainstream view. The levers it lists are real; the trade-offs it highlights are real as well. My concern is that, taken together, this vocabulary turns resilience into a friendly label you can stick on almost any improvement project: more inventory? It is resilience. Less inventory, but in “better” places? Also resilience. A new dashboard? Resilience. A new process? Resilience again.
When a word starts to mean “anything that sounds like a good idea,” it quickly stops being useful.
Why I insist on a sharper boundary
I draw a hard line between the domain of resilience and the domain of ordinary supply chain excellence because the two are governed by different types of knowledge.
Most of what a supply chain team faces is uncertain but not mysterious. Demand varies, but in ways that can be captured – imperfectly, yet usefully – by statistical models. Lead times are noisy, but their variability can be measured. Promotions, price changes, assortment changes, calendar events: they all add structure to this uncertainty. We can attach probabilities and economic consequences to many of these patterns.
In this space, the right question is not “How do we become resilient?” The right question is: “Given what we know about the distributions of demand, lead times, and prices, what is the best decision today in economic terms?” A good decision, in that sense, is a bet: it weighs possible outcomes and their financial impact. It accepts that some days we will lose the bet, but it makes those losses small and affordable.
If we label every failure to do this properly as a “resilience issue,” we excuse a lot of avoidable fragility. A safety stock rule that ignores lead-time uncertainty does not become respectable simply because we say it is part of a resilience strategy. A replenishment process that cannot cope with promotions is not suffering from a resilience failure; it is just poorly designed.
Resilience, as I use the term, only begins where such probabilistic, economics-driven thinking stops being sufficient – where we confront events that sit outside the repertoire of patterns that our models, experience, and data can reasonably cover.
Decisions as bets, and why that matters for shocks
Even when we are not dealing with shocks, every supply chain decision is a bet on the future. We rarely experience it that way, because the decisions are numerous and repetitive: a reorder here, a production batch there, a truck to route, a price to adjust. But behind each of these actions lies an implicit view of what may happen, and of how costly each outcome would be.
What interests me is the shape of that bet.
Many organizations, often unconsciously, design their processes so that decisions are extremely sensitive to a narrow view of the future. A forecast is treated as a single number. Service levels are treated as sacred thresholds. Capacity is run close to saturation. Minimum order quantities and rigid constraints lock in large commitments early. As long as the world behaves roughly as expected, this feels efficient: inventories are low, utilization is high, costs look good.
The moment reality deviates – and it always does, even without a lockdown or a war – these decisions turn out to be brittle. A modest demand surprise, a minor supplier delay, or a small regulatory change propagates through the network in ways that nobody anticipated, because the underlying bets had no room for deviation.
From my perspective, resilience is not primarily about what you do after a shock. It is about the structure of the bets you place before it. A supply chain that systematically makes brittle bets will not magically become resilient when something serious happens. Conversely, a supply chain that habitually prices uncertainty correctly – that accepts some slack where slack is cheap and damaging shortfalls where they are affordable – will often behave gracefully even under pressure.
This is why I see resilience as a side effect of disciplined decision-making under uncertainty, not as a separate layer of processes and dashboards.
Bandwidth, automation, and the human factor
There is another, more human aspect that is often overlooked: the bandwidth of the people who are supposed to care about resilience.
In many companies, supply chain teams live in a state of permanent firefighting. They reconcile inconsistent data from multiple systems, manually override plans that make no sense, jump from one exception to the next, and attend endless meetings to explain yesterday’s problems. The tooling that was supposed to simplify their lives instead generates most of the noise they must sift through.
In such an environment, who has the time – or the mental energy – to think seriously about low-frequency, high-impact shocks? The agenda is fully consumed by problems that are both urgent and self-inflicted.
My position is that any credible resilience agenda starts by freeing up this bandwidth. That means automating the vast majority of routine decisions, not with simplistic rules, but with quantitative engines that understand uncertainty and economics well enough to make thousands of small bets on behalf of the organization. When machines handle what they are good at – repetitive, data-driven choices under stable rules – humans can focus on what they are uniquely suited for: imagining failure modes that have not yet occurred, challenging assumptions, and deciding which structural risks the company is willing to bear.
Without this shift, much of the talk about resilience is just that: talk.
Where I part ways with mainstream practice
All of this creates a quiet but significant divergence between my view of resilience and the one that dominates most industry conversations.
The first divergence concerns forecasting and risk. In mainstream practice, forecasting is usually treated as a separate, almost sacred, activity: a single number per SKU and per period, occasionally seasoned with scenarios. Risk management then arrives on top, with heat maps, registers, and workshops. In my experience, this separation is artificial. Uncertainty is not an add-on; it is the raw material of every decision. When we compress it into single-point forecasts and then paint “risk” around the edges, we are already setting ourselves up for fragile outcomes.
The second divergence concerns metrics. Much of the resilience literature is framed in terms of time-to-recover, minimum acceptable service levels during a disruption, exposure indices, and similar key performance indicators. These can be useful for communication, but if we optimize them directly, we are tempted to treat resilience as a virtue that must be increased in the abstract. I prefer a more prosaic question: for a given class of shock, how much money would we expect to lose under our current design, and how much would it cost to reduce this loss by a certain amount? Once we phrase it that way, resilience stops being mystical. It becomes a capital allocation problem.
The third divergence concerns redundancy. Many resilience playbooks encourage additional inventory, more suppliers, more capacity, and more routes as inherently good things. I do not share this enthusiasm. Some redundancy is extremely valuable; some is pure waste. The difference lies in option value: what does this additional supplier, this spare capacity, or this buffer stock allow us to do in the face of uncertainty that we could not do otherwise, and how often is that option actually likely to be used? Only by answering that question in financial terms do we know whether a given “resilience investment” makes sense.
Finally, there is the question of governance. A lot of mainstream thinking places resilience in committees, frameworks, and certification programs. Those may have their place, but the key decisions that determine resilience are often entrepreneurial in nature: whether to concentrate production in a single very efficient location or to accept the overhead of multiple plants in different jurisdictions; whether to rely on a small number of highly optimized suppliers or to maintain relationships with alternatives that may be less competitive in the short term. These are not technical choices inside the supply chain function; they are strategic decisions about the kinds of shocks the company is willing to live with.
Resilience, robustness, and antifragility
It is also worth distinguishing resilience from two neighboring ideas: robustness and antifragility.
A robust system is one that is barely affected by disturbances within a certain range. It keeps operating much as before. A resilient system suffers when a shock hits, but it recovers. A fragile system is one that cannot recover: the shock pushes it past a point of no return.
Antifragility, a term popularized by Nassim Nicholas Taleb, goes one step further: it describes systems that actually benefit from volatility. They gain from disorder instead of merely surviving it.
In competitive markets, antifragile behavior tends to win in the long run. Companies that treat shocks only as threats will be outcompeted by those that also see them as opportunities: to acquire distressed assets, to shift market share, to renegotiate terms, to accelerate changes that would otherwise take years. Supply chain, by itself, cannot make a company antifragile, but it can either enable or hinder this stance. A network that is constantly on the verge of collapse cannot be opportunistic when disruption strikes.
This is another reason why I resist treating resilience as a narrow technical domain. At some point, the conversation must touch on the fundamental risk appetite and imagination of the firm.
Practical consequences of this view
What does all of this mean in practice?
It means that, before we launch grand “resilience programs,” we should first clean up the basics of how we make decisions under uncertainty. Are we still relying on single-number forecasts and static safety-stock formulas that ignore lead-time variability? Are we enforcing rigid constraints – like arbitrary minimum order quantities or full-truck rules – that wipe out options for the sake of simplicity? Are we measuring performance in ways that reward short-term efficiency illusions and hide long-term fragility?
It means we should invest seriously in decision automation that is worthy of the name: systems that ingest probabilistic views of demand, lead times, and prices, and that optimize for economic outcomes rather than arbitrary fill-rate targets. This is not about buying a dashboard. It is about building or adopting engines that can take responsibility for a large portion of the daily combinatorial workload, so that human experts can focus on structural questions.
It means we should identify the few truly systemic exposures that matter for our business. For each of them, we can ask simple, uncomfortable questions: if this port, this currency, this regulatory environment, or this political region were unavailable to us for a year, what would actually happen? Would we survive, and in what shape? If the honest answer is “we do not know,” then measurement is the first order of business. If the answer is “we would be finished,” then we must decide whether that risk is acceptable. If it is not, the remedy will rarely be a gadget; it will be a structural change.
It also means accepting that resilience has a price. A supply chain that is genuinely more resilient will often look less “efficient” on the very narrow metrics that we have been taught to worship: it may carry more slack, share more margin with partners, or maintain capabilities that look idle most of the time. The question is not whether this price exists, but whether it is worth paying given the shocks we truly care about.
Resilience, as I see it, is not a new layer of complexity to be added on top of an already overloaded discipline. It is the long-term consequence of taking uncertainty and economics seriously in the small decisions we make every day, and of having the courage to make a few big structural choices in full awareness of the shocks we cannot rule out.
If we reserve the word “resilience” for those shocks, and if we treat everything else as ordinary supply chain work that can and should be automated and improved without drama, then the term regains its sharpness. It becomes something we can reason about, invest in, and – when needed – deliberately trade off against other goals.
That, at least, is the kind of resilience I am interested in.
