Before You Fix a Capacity Problem, Make Sure You Actually Have One

A hospital can have an empty bed and still be unable to admit another patient. That sounds contradictory only if capacity is treated as a count of physical assets. In practice, a usable hospital bed requires more than a room and a frame. It has to be set up, staffed, and supported well enough to care for a patient. CDC capacity reporting therefore distinguishes staffed beds from beds that merely exist physically or are licensed on paper.

That distinction captures a larger organizational problem. When demand exceeds what a system can currently handle, the visible symptom is often described as a capacity shortage. Sometimes that diagnosis is correct. But the same symptom can also be produced by the wrong capability mix, blocked allocation, poor configuration, coordination failure, a downstream bottleneck, or an operating model that has been pushed so close to full utilization that it no longer has room for normal variability.

Those conditions do not warrant the same response. Adding resources before understanding which one is present can increase cost while leaving the actual constraint untouched.

Capacity Is a System Property

Organizations often measure capacity by counting resources: employees, servers, hospital beds, factories, machines, hours, or budget. Those measures matter, but they do not necessarily tell us how much useful work the system can actually perform.

Healthcare makes the difference concrete. CDC defines hospital bed capacity in terms of beds that are currently set up, staffed, and able to be used for patients. A physical bed without the staff needed to operate it is not equivalent to a staffed bed. The resource exists, but the operating capability does not.

The same reasoning applies elsewhere. Ten engineers are not interchangeable if the constrained work requires a specialization only two of them possess. A warehouse can have unused floor space while loading docks limit throughput. A computing environment can show idle processors while work is blocked by memory, data movement, permissions, or another constrained component. Capacity is therefore better understood as the ability of the relevant system to meet a particular demand under the conditions in which the work actually has to occur.

The Same Symptom Can Have Different Causes

Southwest Airlines’ December 2022 disruption illustrates how resources can exist but become difficult to deploy. Severe winter weather initially disrupted flight schedules, but the recovery problem grew as cancellations multiplied. Southwest later reported that the volume of unexpected crew changes became too great to address efficiently through its crew-scheduling process. Assignment updates were delayed, which meant the system no longer had sufficiently accurate information to reassign crews effectively. The airline ultimately cancelled more than 16,700 flights between December 21 and December 31, and the U.S. Department of Transportation later described the broader holiday disruption as cancelling 16,900 flights and affecting more than two million passengers.

The relevant lesson is not that Southwest had no crews or aircraft. A crew member who exists but cannot be located accurately in the operating state, connected to the right aircraft, and reassigned within the constraints of the schedule is not usable capacity for the flight that needs coverage. The failure therefore included a configuration and coordination problem. Adding more people into the same broken operating state would not necessarily have restored throughput.

The semiconductor shortages of 2020 and 2021 show a different mechanism. The U.S. Department of Commerce found that shortages were concentrated in particular inputs and applications, including legacy logic chips used in automobiles and medical devices, analog chips, substrates, and wafer-production capacity. Aggregate semiconductor capacity was therefore less informative than the specific type of manufacturing capability that constrained the products customers needed.

Electric transmission provides another variation. The Department of Energy assesses transmission capacity constraints and congestion because generation can exist within the wider grid while physical transmission limits prevent enough electricity from reaching the places that need it. Organizations encounter analogous conditions when capacity exists in one unit but budgets, permissions, interfaces, decision rights, or local ownership prevent it from moving to another. From the point of demand, the condition looks like scarcity. From the perspective of the whole system, the problem may be access or allocation.

Slack Is Not Automatically Waste

Capacity can also be damaged by trying to use too much of it. Managers reasonably look at underutilized resources and ask whether productivity can be improved. The danger is assuming that every period in which a person, machine, or team is not fully occupied represents waste.

Queueing and operations research show why that assumption fails in variable systems. When arrivals and work duration fluctuate, waiting times rise sharply as utilization approaches 100 percent because there is progressively less headroom to absorb even small deviations from the average. A recent Cambridge treatment of operations management describes this as a central utilization-variability tradeoff: near full utilization, minor delays create queues that propagate through the system.

Hospitals make the effect visible. A national study of hospital occupancy and emergency-department boarding found a threshold relationship: when occupancy exceeded 85 percent of staffed beds, boarding exceeded a four-hour standard in 88.9 percent of hospital-months, and median boarding time was much longer than in lower-occupancy periods. The study does not establish 85 percent as a universal target for every system, but it shows why eliminating all apparent headroom can make a variable system less responsive rather than more efficient.

Human capacity adds another layer. John Pencavel’s research on working hours found diminishing returns as hours became long, showing that additional labor time does not translate indefinitely into proportional additional output. A 2024 systematic review and meta-analysis of 85 studies involving more than 288,000 nurses found that burnout was associated with more medication errors, adverse events, missed care, lower safety ratings, and lower assessed quality of care. Those findings do not mean that every highly utilized team is burned out, but they reinforce an important boundary: sustainable human capacity cannot be inferred from the number of hours people can technically remain at work.

Some apparent slack therefore performs a real function. It can provide surge headroom, recovery time, maintenance, learning, problem-solving capacity, and room to absorb the unexpected. Removing it may improve a utilization metric while reducing the system’s effective capacity under the conditions it actually experiences.

Slack Has to Be the Right Slack

The opposite mistake is to romanticize unused resources. Slack is not automatically valuable simply because full utilization can be dangerous. A spare generalist does not solve a shortage of a scarce specialist. Idle capacity in one department may be inaccessible to another. Extra equipment is irrelevant if a different stage is the bottleneck. Resources may also sit unused because work is poorly allocated, duplicated, obsolete, or weakly coordinated.

The useful distinction is between capacity that looks idle because it is performing a buffering or resilience function and capacity that is genuinely unnecessary or badly placed. Protective slack preserves surge response, recovery, maintenance, learning, or reliability. Misplaced slack exists in a form or location that cannot address the constraint. Accidental slack can result from obsolete work, poor allocation, or unnecessary duplication.

This is why utilization cannot be optimized in isolation. The relevant question is how much usable headroom the system needs, where it needs it, and what function that headroom is performing. A more efficient local resource can make the larger system less capable if the optimization removes the buffer that was absorbing variability or preserving quality.

Sometimes the Answer Really Is More Capacity

A careful diagnosis should not become an excuse to avoid genuine scarcity. Organizations sometimes under-resource work and then search indefinitely for process improvements because increasing capacity is expensive.

No scheduling technique can create a qualified specialist who does not exist. Better allocation cannot make a fabrication plant manufacture a chip that requires equipment it does not possess. A transmission line has physical limits. A hospital cannot safely operate a bed without the people and support needed to care for the patient. When the relevant capability is genuinely scarce, adding or developing capacity may be exactly the right answer.

The point is not to resist investment. It is to distinguish a true shortage from the different conditions that merely look like one. The value of diagnosis works in both directions: it can prevent unnecessary expansion, and it can also establish when expansion is genuinely warranted.

Diagnose the Constraint Before Expanding the System

By the time an organization says, ‘We need more capacity,’ several different conditions may already have been compressed into one phrase. Demand may truly exceed the relevant resource. The resource may exist but have the wrong capability. Capacity may be blocked by ownership or access. Resources may be configured badly. Another point in the system may be the actual bottleneck. Or utilization may have been pushed so high that ordinary variability turns every spike into overload.

Those mechanisms can produce the same visible symptoms: work accumulates, deadlines slip, customers wait, and teams appear overwhelmed. The symptom does not establish the cause.

Before expanding the system, it is therefore worth asking what is actually preventing useful work from flowing. Would adding more of the visible resource increase throughput, or would the work simply wait somewhere else? Is the constrained resource the right kind of capability? Does usable capacity exist elsewhere but remain inaccessible? Are approvals, information, sequencing, or dependencies limiting deployment? Does the operating model preserve enough headroom for predictable variability, recovery, and disruption? What evidence would distinguish among those explanations?

The goal is not perfect diagnosis before action. It is enough understanding to avoid an expensive intervention aimed at the wrong mechanism. An organization that mistakes blocked allocation for scarcity can hire people it does not need. One that mistakes a capability gap for a headcount problem can expand the workforce while leaving the missing skill untouched. One that mistakes coordination failure for staffing failure can add complexity to a system already unable to coordinate what it has. And one that removes every pocket of slack in pursuit of utilization can discover that the ‘waste’ it eliminated was the system’s ability to absorb reality.

Capacity is not simply how many resources an organization owns or how busy those resources appear. It is the sustainable ability of the system to meet demand with the right capabilities, in the right configuration, with enough access and enough headroom to remain effective when conditions vary. The practical task is therefore not merely to add capacity. It is to determine which kind of capacity problem the organization actually has before deciding how to fix it.

Selected sources and further reading

• CDC, Hospital Bed Capacity Project and bed definitions - staffed beds as physically available beds with staff on hand. Source

• Janke, Melnick & Venkatesh, JAMA Network Open (2022) - hospital occupancy and emergency-department boarding. Source

• Southwest Airlines 2022 Form 10-K - crew-scheduling and recovery constraints during the December 2022 disruption. Source

• U.S. Department of Transportation - Southwest 2022 holiday disruption and passenger impact. Source

• U.S. Department of Commerce - semiconductor supply-chain bottlenecks by chip type, inputs, and fab capacity. Source

• U.S. Department of Energy, National Transmission Needs Study - transmission capacity constraints and congestion. Source

• Cambridge, Operations Management - utilization, variability, and the rapid growth of waiting near full utilization. Source

• John Pencavel, The Productivity of Working Hours, The Economic Journal (2015). Source

• Li et al., JAMA Network Open (2024) - systematic review and meta-analysis of nurse burnout, patient safety, and quality of care. Source