Rethinking Division of Labor in the Age of Automated Supply Chains
Division of labor is one of those ideas that’s so deeply woven into modern life that we mostly stop seeing it. Adam Smith opens The Wealth of Nations with the now-famous pin factory, where breaking one job into many small tasks multiplies productivity by orders of magnitude.
I share that admiration. Without specialization and trade, there would simply be no “modern” to speak of. Yet, after nearly two decades spent working inside real supply chains, I have also come to believe that we apply the idea of division of labor in ways that quietly work against us—especially inside companies, and even more so inside the software we use to run them.
In my book Introduction to Supply Chain, I argue that supply chain, as a discipline, is about making better decisions on flows of goods under uncertainty, in service of long-term profitability. In this essay, I want to isolate one thread from that larger argument: how division of labor helps us at the global scale, yet often hurts us at the organizational scale, and how automation changes the picture.
The classical story: from pins to global value chains
The classical story is well known. Division of labor, whether within a factory or across countries, explains much of the productivity gains that lifted living standards over the last two centuries. Smith’s pin makers are early cousins of today’s global value chains, where different countries specialize in distinct stages of a product’s life: design in one place, components in another, assembly somewhere else, and distribution on yet another continent.
Modern supply-chain management inherited this narrative. Look at a widely cited framework such as Douglas Lambert’s definition of supply chain management as “the management of relationships in the network of organizations… using key cross-functional business processes to create value.” The emphasis is squarely on coordination across specialized functions and across firms: marketing, logistics, production, purchasing, finance, R&D, and so on.
In that mainstream vision, the job of supply chain management is to ensure that this extended division of labor works smoothly. Functional silos are acknowledged as a problem, but the prescription is almost always more structured collaboration: cross-functional process teams, integrated planning cycles, and in particular Sales & Operations Planning (S&OP). S&OP is usually described as a recurring, cross-functional meeting engine that aligns sales, marketing, supply chain and finance on a single plan.
So far, there is nothing to object to. Division of labor, plus coordination, plus a supporting layer of technology: this is, in broad strokes, how we got here.
But if we pay attention to what actually happens inside companies, especially large ones, a paradox appears.
The hidden cost of internal division of labor
Division of labor is first and foremost a way to cope with human limits. No single person can follow every product, supplier, and customer; so we split the work. One planner takes this region, another that product line, another the long-tail assortment. One department owns purchasing, another pricing, another promotions.
In a lecture titled On Knowledge, Time and Work for Supply Chains, I distinguished between two broad ways of dividing work. One spreads similar activities across many people (“horizontal” division of labor), the other stacks different levels of responsibility in a hierarchy (“vertical” division). Modern companies lean heavily on both.
This works—up to a point. It also creates a series of side effects that are so common we mistake them for laws of nature.
First, complexity translates directly into headcount. Every time a company adds SKUs, channels, regions or constraints, it very often responds by adding planners. The prevailing mental model is roughly linear: twice as many moving parts, twice as many people. It is not unusual to see teams where the daily work consists of scanning through endless lists in spreadsheets or planning tools, making tiny manual adjustments that no one will remember next week.
Second, key levers are fragmented across functions. Pricing and promotions sit in marketing, assortment decisions in merchandising, service promises in sales, while inventory and capacity are left to “supply chain.” Yet all of these decisions shape what moves where and when, and thus shape the economic outcome of the flows. The split is not based on economics; it is based on corporate history.
Third, enterprise systems fossilize yesterday’s division of labor. Most of the logic in ERPs, APS tools, and similar systems is not about economics or statistics; it is about who is allowed to do what, in what order, with which status codes, escalations, and approvals. As I noted recently when discussing product lifecycle management, the vast majority of business logic exists to orchestrate human workflows and handoffs. When we automate decisions properly, a surprising share of this scaffolding becomes redundant.
Fourth, responsibility becomes diluted. When each stage of a process is assigned to a different group, it becomes painfully easy for everyone to be “involved” and yet no one to be answerable for the quality of the final decision. I have seen many “quantitative” initiatives fail, not because the mathematics were flawed, but because the work was sliced into so many pieces—data extraction by IT, data cleaning by one team, forecasting by another, parameter tuning by planners—that no single person or team truly owned the outcome.
This is still division of labor. It still brings local efficiencies. But it does not necessarily help us make better decisions, which is ultimately what matters.
Automation forces a more fundamental split
For roughly forty years, people have talked about automating supply-chain decisions: first inventory control, then distribution requirements planning, and so on. In practice, most companies still rely on people as the primary decision mechanism. Computers provide numbers, dashboards, and alerts; humans remain in the loop at the most granular level.
From my perspective, that is the wrong starting point.
The first and most important division of labor we should draw is not between departments, but between humans and machines.
Machines are extraordinarily good at certain things that dominate modern supply chains: processing vast amounts of transactional data; recomputing decisions daily or even hourly; and sticking to a policy without fatigue or mood swings. Humans, by contrast, are comparatively poor at large-scale repetition but very good at questioning assumptions, interpreting context, and inventing new ways to encode economics into rules.
Once we accept this, the design question changes. Instead of asking “How should we distribute SKUs and suppliers across our planners?”, we should ask, “Which classes of decisions should be fully automated, and how do we design the system that does it?”
In my lectures on quantitative supply chain, I describe the deliverable as a decision engine: an analytical system that turns raw data into concrete decisions such as purchase orders, stock transfers, or price changes, with no manual touch in the daily run. The design of the engine is intensely human work; its routine execution is not.
This is not science fiction. When we insist that the output of the analytical system be actual decisions, not just forecasts or scores, and when we insist that its operation be fully automated, we discover that a great deal of the repetitive planning workload can indeed be mechanized. The result is a different kind of division of labor: a small number of people working on the logic that governs thousands or millions of micro-decisions.
Redrawing roles around the decision engine
If we take this automation-first split seriously, the internal division of labor starts to look very different.
The people who used to spend their days adjusting orders in spreadsheets step into more strategic and investigative roles. Instead of repeatedly asking, “What should I order for this SKU today?”, they ask, “Why did the decision engine recommend this pattern for this family of products?” and “What does this tell us about our costs, constraints, and options?” In effect, they become stewards of the flows, responsible for understanding and refining the economic logic, not for typing numbers.
Meanwhile, a specialized profile emerges at the intersection of supply chain, statistics and software engineering. This person, sometimes called a supply-chain scientist, is accountable for the behavior of the decision engine itself: the way demand uncertainty is modeled, the way shortages and overstocks are economically valued, the way logistical constraints are expressed, and the way all of this is translated into executable code.
Crucially, this scientist does not “belong” to IT, even though they work with code and data. IT retains responsibility for the reliability and security of the data pipelines—for ensuring that transactional systems are mirrored correctly into analytical storage—but the responsibility for shaping decisions sits firmly within the supply-chain function. This explicit split between infrastructure and decision logic is itself a new division of labor, one that preserves clarity of accountability rather than diffusing it.
Finance also re-enters the picture in a more constructive way. Instead of arguing about whether a given forecast is “realistic” in an S&OP meeting, finance and supply chain collaborate to express the company’s actual economic preferences—cost of capital, penalty for stockouts, service commitments to key customers—in a form the decision engine can understand. Once those preferences are encoded, they apply consistently across thousands of decisions, every day, without requiring a meeting each time.
The end result is still specialization. People do not become interchangeable generalists. But the organizing principle is no longer the organigram or the transaction processing sequence; it is the design, operation, and continuous improvement of a decision system.
How this differs from mainstream “integration”
At this point, it is natural to ask: isn’t this just another way of talking about integration? After all, mainstream supply-chain management has spent the last twenty years emphasizing the need to break down silos through cross-functional processes and shared metrics.
There is an important difference.
In the mainstream view, integration means getting more people from more functions into the conversation. The typical S&OP diagram shows sales, marketing, operations, supply chain, and finance around the table, supported by increasingly sophisticated planning software. Collaboration is the scarce resource; technology is there to facilitate it: shared data, shared dashboards, shared workflows.
In my view, integration means something else entirely. It means that the economic logic is unified. The hierarchy of priorities—service vs margin vs capital employed—is expressed once, inside the decision engine, and then applied everywhere. The primary scarce resource is not meeting time but clarity: clarity about what the company is trying to optimize, and about how that intent is translated into operational decisions.
When we start from that angle, we typically discover that many of the coordination mechanisms we built over the years were compensating for missing automation. We needed long meetings because every decision was, in practice, bespoke. We needed elaborate workflows and approval chains because there was no single, trusted, executable policy.
This is why I am skeptical about efforts to “modernize” traditional S&OP by adding a veneer of advanced analytics while keeping its basic structure intact. Whether in debates we host at Lokad or in the academic literature, S&OP is still largely framed as a cross-functional negotiation process, with technology as facilitator. I believe that for many companies, the real step change will come not from better meetings, but from needing far fewer of them in the first place.
Division of labor at the global level: agreement, with a caveat
All of this might sound as if I am opposed to division of labor altogether. I am not. At the global level, I consider deep specialization and trade to be non-negotiable if we care about prosperity. The elaborate international division of labor that characterizes modern value chains is not a fragile curiosity; it is the only reason we can afford the goods and services we currently take for granted.
However, this global specialization comes with systemic fragility, as the recent waves of disruptions have painfully reminded us. When a lockdown closes factories, or a canal gets blocked, or a conflict interrupts exports, the shock propagates through the same networks that normally bring us efficiency. The answer is not to retreat into autarky. It is to become much more precise in how we manage risk: cultivating optionality in suppliers and routes, measuring its cost, and using automation to react quickly when reality deviates from expectation.
In that sense, my disagreement with the mainstream is not about whether division of labor is desirable; it is about where we accept it as given and where we should be willing to redraw it.
A different way to organize work
If we put all the pieces together, a different picture of supply-chain organization emerges.
Globally, we embrace the division of labor that underpins trade and productivity, while being honest about its fragility and deliberate about building options. Inside the firm, we resist the reflex to respond to every increase in complexity with a proportional increase in planners and process layers. Instead, we invest in decision engines that can shoulder the repetitive burden, and we reshape our internal division of labor around the design, governance, and continuous improvement of those engines.
This is not less human. It is more. It treats planners as potential strategists, investigators, and designers of better economic rules, rather than as human middleware between spreadsheets and ERPs. It treats IT as a critical partner in providing solid data infrastructure, without confusing that with ownership of business logic. It treats finance as a co-author of the economic model, not merely as the final approver of budgets.
The classical story of division of labor, from Smith’s pin factory to today’s global value chains, remains valid. But if we stop the story there, we miss the lesson that matters most for contemporary supply chains: in an age where machines can take over much of the repetitive thinking, the truly strategic decision is how we choose to divide labor between humans and machines, and only then how we divide it among ourselves.
