Variation, demand and capacity
Can the system cope with the work—and does the intervention leave room to respond?
Average demand and average capacity can look balanced while people still wait. Work arrives unevenly, jobs take different amounts of time, and the skills or resources needed are not always available together.
Look beyond the averages
Imagine a service with enough time to handle an average day’s orders. Requests cluster in the morning, some need specialist help, and urgent work interrupts planned work. A quiet afternoon cannot undo every delay created earlier.
- Amount: how much work arrives, including rework and unmet demand?
- Timing: when does it arrive, how long does it take and when is capacity available?
- Variety: what different needs, skills and responses are involved?
Follow the work with the people doing it. Check these patterns over time and at the steps where queues form—not just in organisation-wide totals. Find what is limiting the result.
Keep capacity to respond
Utilisation is the proportion of available capacity being used. Slack is capacity not already committed, which can help absorb variation and recover from disruption.
Busy is not the same as responsive. When arrivals and job durations vary, work can queue even when average capacity exceeds average demand. As utilisation approaches full capacity, there is less room to absorb that variation, so waiting can rise sharply. Reducing avoidable variation and keeping some capacity to respond can both help.
How to read the graph—and its limits
In the higher-variability example, average queue waiting rises from four times the average job duration at 80% utilisation to nine times at 90%. This is a gradual, increasingly steep rise—not a fixed cliff or an exponential curve.
The graph uses Kingman’s approximation for a single resource serving one job at a time, in steady state, with first-come-first-served work and no abandonment or loss. Both curves keep the same average job duration. The variability factors are 1 and 0.25: the coefficients of variation of both interarrival times and job durations are respectively 1 and 0.5.
The model is shown only up to 95% utilisation. As utilisation approaches 100%, its predicted average waiting grows without bound. Real services may respond through overtime, changed priorities, lost demand or other changes; pooled teams and different queue arrangements need appropriate models.
This is a waiting-time graph, not a throughput graph. Little’s Law connects average work in the system, throughput and time; it does not by itself give this waiting curve. Use local evidence and suitable analysis to choose capacity—not these illustrative values as targets.
Around 80% utilisation can be a useful planning prompt—not a universal limit. There is no fixed cliff at 80%, and 20% slack is not a rule for every resource. The margin needed depends on variability, pooling of resources and the consequences of delay.
Protect useful flexibility where and when it is needed, with the right skills. Do not fill every gap simply to improve a utilisation score—or assume spare capacity in one team can relieve a constraint elsewhere.
Design and test the arrangement
- Reduce avoidable variation. Remove rework, missing information and artificial batches; standardise routine work without ignoring genuine differences in need.
- Match capacity to demand. Consider flexible schedules, shared capacity or cross-training where safe and practical. Protect a clear route for urgent work and exceptions.
- Keep a deliberate margin. Agree how response capacity will be protected and used. If demand persistently exceeds capacity, slack alone is not a solution: address capacity, scope or demand with those affected.
- Check the whole result. Monitor completed work, waiting, quality, safety, workload and cost. Did the change improve flow or simply transfer the queue or burden?
How Ashby’s Law helps explain this
Ashby’s Law of Requisite Variety concerns the range of responses needed to manage the range of disturbances that matter to a system’s goal.
For a service, a useful question is: can our arrangements respond to the different needs we actually encounter? Reduce avoidable complexity and build appropriate ways to handle what remains—for example, a clear routine route plus skilled judgement for exceptions.
Variety is not the same as numerical variability or workload. Extra hours alone may not help if the required skills, information or authority are missing. Nor does the law mean every difference should be standardised away.
Sources and further reading
W. Ross Ashby, An Introduction to Cybernetics (1956), Chapter 11: requisite variety.
Start here: the graph and explanation on this page introduce why variable work creates queues near full utilisation.
A short external reading: Christopher A. Bain and colleagues, Myths of ideal hospital occupancy (2010), a two-page article explaining why there is no single ideal occupancy percentage. It uses hospitals as its example; it is not a general staffing formula.
What’s your next useful question?
Where does the work need more flexibility—not simply more activity?
Return to Right problem? →Return to Right intervention? →