Capacity Utilization and Efficiency: When Busy Isn't Better
Key takeaways
- Capacity utilization is the share of available capacity a resource uses, while efficiency is good output produced per unit of that capacity.
- Pushing non-bottleneck resources toward high utilization builds WIP and lengthens lead time without adding system throughput.
- Only the bottleneck resource should run near full utilization; non-bottlenecks should deliberately run below it, paced to the constraint.
- The Federal Reserve's G.17 release benchmarks US capacity utilization against estimated normal capacity, not a theoretical maximum.
- Set utilization targets per resource, not per site, and re-test where the constraint sits at each demand level before quoting client work.
Capacity utilization and efficiency measure different things: utilization is the share of available capacity you actually use, while efficiency is output produced per unit of that capacity. Pushing utilization up everywhere often lowers system efficiency, because non-bottleneck resources running flat out just pile up WIP without adding throughput. This article shows how the two measures relate, how global and US capacity utilization figures are built, and how to set targets per resource when scoping client work.
What is the difference between capacity utilization and efficiency?
Capacity utilization is the share of available capacity a resource actually uses: hours worked divided by hours available. Efficiency is output per unit of input: good units produced per hour, per person, or per dollar of input. A machine running at 95% utilization can still be inefficient if 20% of that time produces defects or rework.
The distinction matters because the two can move in opposite directions. Pushing a non-bottleneck resource to higher utilization typically builds WIP and lengthens lead time, so total system output per input falls even as that resource looks busier. The Federal Reserve's G.17 release defines US capacity utilization along these lines and publishes the underlying data monthly.
When scoping client work, measure both separately. The process capacity calculator separates the two so a busy-looking resource does not hide a throughput problem.
How do global and US capacity utilization figures get measured?
The US headline number comes from the Federal Reserve's G.17 release, published monthly. It reports capacity utilization for the industrial sector: manufacturing, mining, utilities, and selected industries. Global figures are assembled from comparable national statistical sources, so definitions and coverage vary by country.
What normal capacity utilization means in the Fed's series
The Fed measures output against an estimated "normal" capacity level, not the theoretical maximum. Normal capacity reflects sustainable full-operation rates, allowing for typical downtime, maintenance, and product-mix shifts. That is why the published figure can approach but rarely reaches 100%. When you benchmark a client's utilization against the headline number, you are comparing against a similarly adjusted baseline, not a physical ceiling.
Why the headline number hides sector-level spread
The aggregate masks wide variation: one industry can run near full capacity while another sits far below, yet the average looks moderate. The same problem appears in a single client site, where a support team's utilization figure can hide one overloaded queue and several underused ones (see the support-ticket capacity example). Before quoting engagement work, pull the sector-level series or the resource-level data, not just the headline.
Why does pushing utilization up often push efficiency down?
Utilization measures how busy a resource is. Efficiency measures how much good output the system ships per unit of input. Pushing every resource toward 100% busy raises the first number while lowering the second, because busy-ness creates queues.
The queueing effect in plain numbers
Queueing math is unforgiving: as utilization approaches 100%, waiting time grows non-linearly, not linearly. A resource at 90% busy already builds a backlog most days; at 98%, any hiccup becomes a queue that never drains. That queue is lead time, and lead time is what your client's customer feels. The Fed's G.17 notes define normal capacity utilization against a normal, sustainable operating rate, precisely because running flat-out is not sustainable output.
Where WIP hides the problem
High utilization on non-bottlenecks converts into WIP sitting between processes. WIP hides defects, delays feedback, and inflates lead time while the utilization dashboard still looks green. The repair workshop example shows this pattern: every station busy, throughput flat. If the constraint is downstream, upstream busy-ness is waste with a good-looking metric attached.
What does the theory of constraints say about bottleneck vs non-bottleneck resources?
Theory of constraints draws a hard line between the two. The bottleneck sets system capacity, so an hour lost there is an hour lost for the entire process. No amount of local improvement elsewhere recovers it. A non-bottleneck is different. An hour saved there is a mirage: the resource already has spare capacity, so speeding it up only builds more WIP in front of the constraint.
Goldratt's rule is that the bottleneck should run at or near full utilization while non-bottlenecks deliberately run below it, paced to the constraint's throughput rather than to their own maximum. This is why a single utilization target across every resource is a scoping error. Before quoting, separate resources by type: protect the constraint, and let everything else idle on purpose. The practical consequence is that "keep everyone busy" is not a neutral default but an active choice to trade system efficiency for local busy-ness, and it is the most common scoping mistake in capacity work. If you want to see how this plays out on a production line, the worked example at the manufacturing-line capacity page walks through it.
Worked example: one line, two resources, two utilization targets
Take a two-step process. Step A (the bottleneck) can produce 10 units/hour; Step B can produce 18 units/hour. Demand is 9 units/hour. Run the bottleneck at 90% utilization (9 units/hour) and Step B only needs 9/18 = 50% utilization to keep up. For context on how utilization levels are benchmarked across economies, the World Bank publishes comparable indicator data through its open data portal.
What happens if you balance both at 90%
Pushing both steps to 90% changes nothing about output: the bottleneck still caps the line at 9 units/hour. It only fills Step B with 7.2 units/hour of excess WIP (16.2 produced minus 9 consumed), inflating lead time and hiding defects. Same throughput, worse cash conversion, longer lead time. That is the naive 90/90 target failing.
| Target | Bottleneck util. | Non-bottleneck util. | Throughput | Approx. WIP | Lead time |
|---|---|---|---|---|---|
| 90/90 | 90% | 90% | 9/hr | 7.2/hr excess | High |
| 90/55 | 90% | 55% | 9/hr | Near zero | Low |
The 90/55 line matches output with a fraction of the WIP. We walk through the same arithmetic on a service process in our invoice-processing capacity example.
Where the constraint moves when demand changes
If demand rises to 11 units/hour, Step A becomes the binding constraint at 110% of capacity, and the fix shifts to Step A: offload work, add a shift, or raise its cycle-time performance. If demand falls to 7, Step B's target drops to roughly 39%, and the constraint question becomes commercial, not physical. Scoping should therefore re-test the constraint location at each demand level before quoting work.
How should you set utilization targets when scoping client work?
Measure capacity utilization per resource, not per site. A plant-level figure hides the one machine or step that governs system throughput, and quoting against an average is how engagements get scoped around the wrong constraint.
Classify each resource before setting any target. The bottleneck should run near full utilization, because idle time there is lost system output. Non-bottleneck resources should sit deliberately below it: running them hot just builds WIP the constraint cannot absorb, lengthening lead time without adding throughput.
A five-question scoping checklist
- Which single resource sets system capacity, and how do you know?
- What is current utilization for that resource, measured over a representative week?
- How much WIP sits in front of it, and how long is the backlog?
- What is the demand gap between what clients ask for and current throughput?
- If the constraint moves, which resource becomes the next bottleneck?
Answer those five before quoting. The Constraint Impact Calculator on Improve First is built for exactly this scoping step: identify the limiting resource, test whether an improvement lifts whole-system output, and see where the constraint moves next. Where it falls short: it models one process at a time and will not replace a full simulation for multi-line plants, so treat it as a scoping tool, not the engagement itself.
Which tools can test utilization vs efficiency trade-offs?
Criteria for a quick scoping tool
A scoping tool has to do three things fast: model a multi-resource process, show what happens to lead time when utilization on a specific resource rises, and flag where the constraint moves next. Anything that cannot show the utilization-vs-lead-time relationship is decoration, because that relationship is the whole trade-off. Queueing theory makes the stakes concrete: as utilization approaches 100%, waiting time grows nonlinearly, per Kingman's formula.
Where each option falls short
A spreadsheet capacity model (Excel or Google Sheets) is the status quo: free, transparent, and already on the client's machine, but it struggles to show queueing effects or constraint migration without heavy manual formula work. ProcessModel simulates full processes properly, though it is a deeper tool than a pre-quote scoping pass needs. DO.Impact and FlowVisual are lighter-weight direct alternatives. KaiNexus sits adjacent, managing improvement work rather than modelling the constraint itself.
Improve First's Constraint Impact Calculator fits the scoping brief directly: it identifies the limiting resource, tests whether an improvement shifts system throughput, and predicts where the constraint moves. Where it falls short: it is a scoping tool, not a simulator, so it will not model detailed variability, shift patterns, or batch logic the way ProcessModel can. If the engagement wins on the back of the scoping report, you will still need deeper simulation for the full diagnostic.
How can you track capacity utilization yourself over time?
For the external benchmark, the Federal Reserve's G.17 release publishes US capacity utilization monthly, covering total industry plus manufacturing, mining, and utilities breakdowns. Watch the trend, not the single month: a falling figure changes how much headroom your client's demand gap really has.
Internal tracking matters more for scoping. Log cycle time, WIP, and throughput at each resource weekly, and calculate utilization as run time divided by available time. One spreadsheet column per resource is enough to start.
Reconcile the two by using the Fed figure as context and your own constraint data as the decision input. The bottleneck resource should sit near high utilization; every non-bottleneck should show deliberate slack. When the constraint moves, your tracker shows it before the client feels it in lead time.
FAQ
Is 100% capacity utilization a good thing?
Almost never. As utilization approaches 100%, waiting time grows non-linearly, so any hiccup creates a backlog that never drains. Only the bottleneck resource should run near full, because idle time there is lost system output. Non-bottlenecks pushed to 100% just convert spare capacity into WIP, lengthening lead time while throughput stays flat.
What is a normal capacity utilization rate for manufacturing?
The Federal Reserve's G.17 series measures output against an estimated "normal" capacity level, not a theoretical maximum. That baseline allows for typical downtime, maintenance, and product-mix shifts, which is why the published figure approaches but rarely reaches 100%. Treat the headline number as an adjusted benchmark, then pull sector-level or resource-level data before comparing a client.
Can utilization be high while productivity is low?
Yes. A resource at 95% utilization can spend a fifth of that time on defects or rework, so good output per hour stays low while the dashboard looks green. High utilization on non-bottlenecks also piles WIP in front of the constraint, hiding defects and delaying feedback. Measure the two separately, per resource, not per site.
How do you calculate capacity utilization in Excel or Google Sheets?
One column per resource is enough to start. Log available hours and run time weekly, then divide run time by available time for utilization. Add columns for WIP and throughput so you can see whether extra busy-ness produces more shipped output or just a longer queue in front of the constraint. Track the trend, not a single week.
What utilization rate should a service business aim for?
There is no single number. The constraint resource should run near full utilization; everything else should deliberately sit below it, paced to the constraint's throughput. In the two-step example, the non-bottleneck ran at 55% while matching full output. Measure utilization per team or queue, since one average can hide an overloaded queue beside several underused ones.