Kingman's Equation: Why High Utilization Destroys Flow

The VUT equation explained: why queues explode as utilization nears 100%, how variability multiplies waiting time, with three worked scenarios.

20 min read 8 sections Guide 3 of 16, The Process Architect's Blueprint Updated

UTILIZATION →WAIT
Kingman's equationWait ≈ variability × u ÷ (1 − u) × process time

New to this topic?

Read these first to get the most out of this guide:

VUT
The Three Drivers of Queue Time
19×
Queue Multiplier at 95% Utilization
~70%
Queue Reduction from 95%→85%
≤85%
Target Utilization (High-V Systems)
Where the knee sits Variability does not move you along the queueing curve; it picks which curve you are on. One working shift of queue arrives at 96.6% utilization on the low-variability machine and at 53.7% on the high-variability one.

Both machines are Worked Example 2, plotted from its own figures. Wait = V × u/(1−u) × Te, so each machine is u/(1−u) scaled by a single constant k = V × Te. Machine A: V = (0.40² + 0.35²)/2 = 0.141, Te = 2.0 h, k = 0.283 h. Machine B: V = (1.30² + 1.50²)/2 = 1.97, Te = 3.5 h, k = 6.895 h, a 24× difference. The marked points are the article's: A at 92% is 0.283 × 11.50 = 3.2 h; B at 78% is 6.895 × 3.55 = 24.5 h. The crossings solve 8 = k × u/(1−u): u = 96.6% for A, u = 53.7% for B. The y axis stops at 30 h so Machine A's full curve fits (28.0 h at 99%); Machine B leaves the top at 81.3% and reaches 131 h at 95%, so the cap understates the gap rather than flattering it.

The Equation That Explains Every Long Queue in Your Facility

If Little’s Law tells you what your lead time is, Kingman’s Equation tells you why. Specifically, it explains why queues form in front of machines and workstations, how long jobs will wait in those queues, and what drives the explosive growth in wait time that most managers have experienced but never understood mathematically.

WIP 7LEAD TIME
Little's lawWIP = throughput × lead timeThe Little's law guide

The equation, developed by Sir John Kingman in 1961 (the same year as Little’s proof, a remarkable year for production science), is known as the VUT equation because it has three terms:

Kingman’s Equation (The VUT Equation)

Expected Wait Time ~ V × U × T

Where:

V = Variability Factor = (Ca² + Ce²) / 2

U = Utilization Factor = u / (1 − u), where u = arrival rate × mean process time

T = Mean effective process time (Te)

Let’s define each component precisely:

ComponentSymbolDefinitionTypical Aerospace Range
Coefficient of variation (arrivals) Ca Standard deviation of inter-arrival times divided by the mean inter-arrival time. Measures how irregular the arrival pattern is. Ca = 0 means perfectly regular arrivals; Ca = 1 means random (Poisson) arrivals; Ca > 1 means highly irregular. 0.5 to 1.5 (aerospace is typically high due to batch releases, rework re-entries, and variable upstream output)
Coefficient of variation (process time) Ce Standard deviation of effective process times divided by the mean effective process time. Measures how variable the processing itself is. Includes setup variation, tool changes, quality holds, and operator method differences. 0.5 to 2.0 (aerospace is high due to complex setups, engineering changes, inspection requirements, and non-repetitive work content)
Utilization u The fraction of time the resource is busy. Calculated as: mean arrival rate × mean effective process time. If a machine processes jobs that arrive every 30 minutes on average and each job takes 27 minutes on average, u = 27/30 = 0.90. 0.70 to 0.98 (many aerospace shops push constraints to 95%+)
Mean effective process time Te The average time to process one unit, including all sources of delay: setup, processing, inspection, minor stops, and detectable quality rework. “Effective” means the real average, not the engineered standard. Varies widely: 0.5 hrs (simple turning) to 40+ hrs (complex 5-axis programs)

The Utilization Multiplier: Where the Explosion Happens

The U term (u / (1 − u)) is the engine of queue explosion. This term is what makes Kingman’s Equation non-linear and counterintuitive. Most managers think of utilization as a linear relationship: 90% utilized should mean 10% more queue than 80% utilized. It does not. The relationship is hyperbolic.

M1M2M3M4
Production scheduleJobs slot onto machine lanes in time; the due-date line sweeps acrossThe Production schedule guide
Utilization (u)Utilization Factor: u/(1−u)Relative Queue Time (vs. 50%)What It Feels Like
50%1.01.0×Short queues, machine often idle, management panics about “wasted capacity”
60%1.51.5×Manageable queues, occasional short waits
70%2.32.3×Noticeable queues forming, still manageable
75%3.03.0×Queues are a regular presence
80%4.04.0×Queues are a constant feature, schedule adherence starts to slip
85%5.75.7×Significant queues, frequent expediting required
90%9.09.0×Long queues, chronic schedule misses, expediting is a full-time job
92%11.511.5×Production control is mostly firefighting
95%19.019.0×Multi-week queues, schedule is fiction, management demands more overtime
97%32.332.3×Queue dominates the value stream, parts “disappear” into the system
99%99.099.0×System is effectively gridlocked

Read that table again. Going from 85% to 95% utilization (a mere 10 percentage points) increases the utilization factor from 5.7 to 19.0. That is a 3.3× increase in queue time for a 10-point change in utilization. Going from 90% to 99% (a 9-point increase) multiplies queue time by 11×.

This is the mathematical explanation for a phenomenon every production leader has observed but few can explain: the machine that runs at 97% utilization and has a three-week queue, right next to the machine that runs at 80% utilization and has a two-day queue. It is not random. It is not bad luck. It is Kingman’s Equation.

The Machine You Are Proudest Of Is Probably Destroying Your Schedule

The machine that management is proudest of running at 97% utilization is probably the machine that is destroying your master schedule. At 97% utilization, the utilization factor is 32.3, meaning queue time is 32 times what it would be at 50% utilization. That “highly efficient” machine is generating weeks of lead time that are invisible in the utilization report but devastatingly visible in your delivery performance.

Why Variability in Aerospace Is Structurally High

The V term, (Ca² + Ce²) / 2, is the variability multiplier. When V = 0.5 (low variability), queue behavior is relatively benign even at moderate utilization. When V = 2.0 (high variability), queues form aggressively at any utilization above 70%.

In aerospace manufacturing, variability is structurally high. This is not a temporary condition that can be engineered away. It is inherent to the nature of the work:

Variability SourceAffectsMechanismTypical Impact on Ce
Engineering changesProcess timeMid-production ECNs change work content, tooling requirements, and inspection criteria. A part that took 4 hours last month may take 6 hours this month due to a design revision.+0.2 to 0.5
Complex setupsProcess time5-axis CNC programs with 15+ tool changes, custom fixturing, and first-article verification. Setup time can vary from 45 minutes to 3 hours depending on the part family and fixture condition.+0.3 to 0.6
Raw material variationProcess timeForging quality, casting porosity, composite pre-preg age: all affect machining parameters, layup behavior, and rework probability.+0.1 to 0.3
Inspection holdsBoth arrival and processFirst-article inspection, in-process verification, NDT requirements. Hold times are highly variable, from hours to days depending on inspector availability and finding disposition.+0.3 to 0.8
Rework re-entriesArrival patternParts that fail inspection re-enter the queue unpredictably, creating arrival spikes that disrupt the flow pattern.Increases Ca by 0.2 to 0.5
Non-repetitive work contentProcess timeHigh-mix aerospace shops may run 50+ different part numbers through the same machine. Each has different cycle times, tooling, and programs.+0.3 to 0.7

The combined effect puts typical aerospace V values in the range of 1.0 to 3.0. Compare this to a high-volume automotive stamping operation where V might be 0.3 to 0.5. This is why utilization targets that work perfectly in automotive (92 to 95%) are catastrophic in aerospace. The safe utilization level depends entirely on your variability.

Variability Determines Whether Utilization Is Safe

Never quote a utilization target without knowing your V. A machine with V = 0.5 can safely run at 90% utilization. The same machine with V = 2.0 will have queue times 4 times longer at the same utilization. Variability is not a secondary factor. It is the factor that determines how much capacity buffer your system needs to maintain flow.

The Counterintuitive Truth: Reducing Utilization Is the Only Lever

In a high-variability environment, you have three theoretical levers to reduce queue time: reduce V, reduce U, or reduce T. In practice, only one is immediately actionable.

The constraintOutput is set by the constraint; an hour lost there is lost for the systemThe constraint guide

Reduce V (variability): This requires eliminating engineering changes, standardizing setups, reducing rework, and smoothing arrival patterns. All worthy goals, and all take months or years of sustained effort. V reduction is a long-term strategy, not an immediate fix.

Reduce T (process time): This means speeding up the machine or finding a faster process. In aerospace, process times are often constrained by physics (cutting speeds for titanium, cure times for composites, NDT scan rates) and cannot be significantly reduced without capital investment in new technology.

Reduce U (utilization): This is the only lever that works immediately and is within management’s direct control. Reducing utilization means reducing the load on the machine, which means either adding capacity (a second machine, a second shift) or reducing the arrival rate (sending some work to an alternative resource, an outside vendor, or simply scheduling fewer jobs through the bottleneck).

Here is the counterintuitive part: reducing utilization does not reduce output. If the machine is not the system constraint, reducing its utilization has zero impact on facility throughput (the constraint determines throughput, see Guide 06). If the machine is the constraint, then the goal is to maximize its effective throughput, which may mean running it at slightly lower utilization with faster changeovers and fewer interruptions, resulting in more actual output with less queue time.

Three Aerospace Scenarios: The VUT Equation in Action

4 ERP3 MES2 CONTROL1 SENSE0 PROCESS
ISA-95 levelsFive levels stack from sensors at the bottom to the business system at the top; orders go down and results come up
Worked Example 1: 5-Axis CNC Machining Center Make Shop

Scenario: A 5-axis CNC machining center processes structural titanium components. Given data:

ParameterValue
Mean effective process time (Te)4.0 hours
Ca (arrival variability)0.85
Ce (process variability)1.10

Step 1: Calculate V.

V = (Ca² + Ce²) / 2 = (0.85² + 1.10²) / 2 = (0.7225 + 1.21) / 2 = 1.9325 / 2 = 0.966

Step 2: Calculate wait time at multiple utilization levels.

Wait Time = V × (u / (1 − u)) × Te

Utilizationu/(1−u)Wait Time CalculationExpected Queue WaitTotal Time (Wait + Process)
85%5.670.966 × 5.67 × 4.021.9 hours25.9 hours (~3.2 days)
90%9.000.966 × 9.00 × 4.034.8 hours38.8 hours (~4.8 days)
95%19.000.966 × 19.00 × 4.073.4 hours77.4 hours (~9.7 days)
99%99.000.966 × 99.00 × 4.0382.5 hours386.5 hours (~48.3 days)

Interpretation: At 85% utilization, a part waits about 22 hours in queue, roughly one working day. At 95%, the same part waits 73 hours, over 9 working days just in queue, for a 4-hour process. At 99%, the queue wait is nearly 48 working days. The process time hasn’t changed. The variability hasn’t changed. The only difference is 14 percentage points of utilization, and queue time has increased by 17×.

Worked Example 2: Low Variability vs. High Variability Counterintuitive Comparison

Scenario: Two machines in the same facility. Which has the longer queue?

ParameterMachine AMachine B
Utilization92%78%
Ca0.401.30
Ce0.351.50
Te2.0 hours3.5 hours

Machine A:

V = (0.40² + 0.35²) / 2 = (0.16 + 0.1225) / 2 = 0.141

U = 0.92 / (1 − 0.92) = 0.92 / 0.08 = 11.5

Wait = 0.141 × 11.5 × 2.0 = 3.24 hours

Machine B:

V = (1.30² + 1.50²) / 2 = (1.69 + 2.25) / 2 = 1.97

U = 0.78 / (1 − 0.78) = 0.78 / 0.22 = 3.55

Wait = 1.97 × 3.55 × 3.5 = 24.5 hours

Result: Machine B has a queue 7.5× longer than Machine A, despite running at 14 points lower utilization.

Interpretation: This is the result that breaks most managers’ mental models. Machine A is running “hotter” but has low variability: it processes similar parts with consistent setups. Machine B is running at a seemingly comfortable 78% but has massive variability from mixed part families, complex setups, and frequent rework re-entries. The variability factor (V = 1.97 vs. 0.141, a 14× difference) overwhelms the utilization advantage.

The lesson: You cannot assess queue risk from utilization alone. You must know V.

The picture above is why the arithmetic came out that way. Machine A and Machine B are not two points on one curve. They are on two different curves, because V × Te is the constant that scales the whole thing. Every machine has the same knee; variability decides where along the utilization axis it falls. Machine B’s knee has already been passed at 54%. Machine A’s does not arrive until 97%. That is the entire practical content of Kingman’s Equation: a utilization target is meaningless until you know which curve you are standing on.

Worked Example 3: The 10-Point Utilization Reduction The Biggest Lever

Scenario: Your constraint machining center runs at 95% utilization with moderate-high variability. You are considering adding a second shift on Saturdays to reduce utilization to 85%. What is the impact on queue time?

Given: V = 1.2, Te = 5.0 hours

At 95% utilization:

Wait = 1.2 × (0.95 / 0.05) × 5.0 = 1.2 × 19.0 × 5.0 = 114.0 hours (14.3 working days)

At 85% utilization:

Wait = 1.2 × (0.85 / 0.15) × 5.0 = 1.2 × 5.67 × 5.0 = 34.0 hours (4.3 working days)

Reduction: 114.0 − 34.0 = 80.0 hours saved per part. That is a 70.2% reduction in queue wait time from a 10-percentage-point reduction in utilization.

Interpretation: Adding Saturday shifts does not just “add a little buffer.” It collapses the queue from over two weeks to under one week. Every part through this machine ships 10 working days sooner. If this machine processes 15 parts per week, that is 150 part-days of lead time eliminated every week, which cascades through the entire downstream value stream.

What management usually does instead: Runs the machine 7 days a week at 99% utilization, which mathematically makes queue times even worse, not better, because the arrival rate increases to match the expanded capacity while variability remains unchanged.

How to Estimate Your Variability Without Statistical Software

You do not need Minitab or JMP to get a useful estimate of V. Here is the practical field method:

STEP · KEY POINT · WHY
Standard workThe current best way, written down, so it can be improvedThe Standard work guide

Collect 20 to 30 observations of actual process time

Record the actual time to process 20 to 30 consecutive jobs through the workstation. Include setup time, processing time, and any in-station delays. Use a logbook at the machine, not ERP data (ERP often records start/end timestamps that include queue time).

Calculate the mean and standard deviation

Mean = sum of all observations ÷ count. Standard deviation = use a spreadsheet or calculator. These are basic descriptive statistics.

Calculate Ce

Ce = standard deviation ÷ mean. If your mean process time is 4.0 hours and the standard deviation is 2.8 hours, Ce = 2.8 / 4.0 = 0.70.

Repeat for arrival times to get Ca

Record the time between consecutive job arrivals at the workstation. Calculate mean, standard deviation, and Ca = std dev ÷ mean.

Use the quick classification if you cannot collect data

Low variability (V ~ 0.3 to 0.5): Repetitive work, standard setups, stable arrivals. Medium variability (V ~ 0.5 to 1.5): Mixed parts, variable setups, some rework. High variability (V ~ 1.5 to 3.0): High-mix, complex setups, frequent rework, engineering changes. Most aerospace machine shops fall in the medium-to-high range.

Exploiting the Constraint from a VUT Perspective

The Theory of Constraints says to “exploit the constraint”: get maximum output from the bottleneck. VUT tells you exactly what that means mathematically: reduce the variability at the constraint while maintaining its throughput.

Takt timeTakt = net available time ÷ customer demandThe Takt time guide

Here is the difference between “exploiting” the constraint and “running it harder”:

Exploiting the Constraint (VUT-Informed)

  • Reduce Ce through setup reduction (SMED), preventive maintenance, and quality at the source
  • Smooth Ca by controlling the release rate upstream (CONWIP)
  • Pre-stage tooling, programs, and fixtures so the constraint never waits for support
  • Run the constraint through breaks by rotating operators
  • Move inspection and rework off the constraint to a parallel resource
  • Result: same or higher TH with lower queue time

Running It Harder (VUT-Ignorant)

  • Push utilization to 99% by releasing more work orders into the queue
  • Skip preventive maintenance to “keep the machine running”
  • Rush setups, creating more process variability and quality issues
  • Override scheduling priorities with daily expedites, increasing arrival variability
  • Accept rework at the constraint to “keep it busy”
  • Result: same or lower TH with exponentially higher queue time

Every Metric That Rewards High Utilization Conflicts with Flow

Every management metric that rewards high machine utilization is in direct mathematical conflict with flow. OEE targets of 95%+, utilization dashboards with green/yellow/red at 90/80/70, shift reports that highlight “idle time”: all of these incentivize behavior that Kingman’s Equation proves will increase queue times. The Process Architect must either change the metrics or educate leadership on why the metrics are counterproductive.

Strategic Idle Time Is Not Waste: It Is the System’s Immune System

Capacity buffer (the deliberate maintenance of some idle time at non-constraint resources) is the system’s ability to absorb variability shocks without queue explosion. When a machine upstream of the constraint breaks down for 2 hours, the buffer capacity at other machines allows the system to catch up without creating a cascade of queue growth. Eliminating all idle time eliminates the system’s ability to recover from disruption.

The Bottom Line

Kingman’s Equation gives you the quantitative framework to understand queue behavior in your facility. V, U, and T (variability, utilization, and process time) are the three levers. In aerospace, where V is structurally high, the most powerful immediate lever is U: reducing utilization at critical resources. This does not mean accepting waste. It means understanding that strategic capacity buffer is the price of flow, and flow is the mechanism that delivers parts on time. The next guide, Takt Time, applies these physics to the Assembly Shop, where the heartbeat of production is set by customer demand, not machine capacity.

Interactive Demo

Pull the three levers. Set utilization, arrival and process variability, and process time, and watch the queue (not the process) decide the lead time.

MAKEUSE
Push and pullPush runs to a schedule; pull runs to a signal from downstreamThe Push and pull guide

Queue Time Explorer (VUT)

Wait time = V × (u ÷ (1 − u)) × Te. Push utilization toward 100% and watch the queue, not the process, decide your lead time. Defaults are the page's 5-axis CNC worked example.

95%
5099
0.85
0.13
1.10
0.13
4.0 h
0.512
0h46h92h85% target↑ off the chart50%60%70%80%90%99%Utilization
0.97
V
19.0
u ÷ (1 − u)
73.4 h
Queue wait
9.7 d
Total time
The same machine at 85%: 21.9 hours of queue instead of 73.4 — 70% less waiting. Nothing about the process changed. Only the capacity buffer did.

Frequently asked questions

What is Kingman's equation in manufacturing?

Kingman's equation (also called the VUT equation) calculates the expected waiting time in a production queue: Wait Time ~ V × (U / 1-U) × T, where V is the variability factor, U is utilization, and T is the mean process time. It mathematically proves that queue times grow exponentially as utilization approaches 100%, especially when variability is high.

Why does high machine utilization cause long queues?

Because of the U/(1-U) term in Kingman's equation. At 50% utilization, this multiplier is 1.0. At 90%, it is 9.0. At 95%, it is 19.0. At 99%, it is 99.0. Queue time grows exponentially, not linearly, as utilization increases. The closer you push a machine to 100%, the longer every job waits, and the relationship accelerates dramatically above 85%.

How do you calculate waiting time in a production queue?

Use Kingman's equation: Wait Time ~ ((Ca² + Ce²) / 2) × (U / (1 − U)) × Te, where Ca is the coefficient of variation of inter-arrival times, Ce is the coefficient of variation of process times, U is utilization, and Te is mean effective process time. Estimate Ca and Ce from observed data, calculate U from arrival rate × process time, and multiply the three terms.

What is the relationship between utilization and lead time?

Through Kingman's equation, utilization has an exponential relationship with waiting time (and therefore lead time). Below 70% utilization, queue times are manageable. Between 70-85%, they grow noticeably. Above 85%, they accelerate sharply. Above 95%, they explode. This is why a machine running at 97% utilization often has a 3-week queue while a machine at 80% has a 2-day queue.

Share this guide: LinkedIn X Follow SymplProcess on LinkedIn

Take this to a room

The running order

For anyone who sets or defends utilisation targets. They should leave able to explain why a 95 per cent utilised machine generates weeks of queue, and what V is on their own line.

7 beats · 14 min
  1. 1

    Three things make a queue

    Variability, utilisation, and process time. Multiply them and you have the wait.

    • V is the variability term: how erratic arrivals and process times are.
    • U is the utilisation term, and it is not linear.
    • T is the process time itself.
    • Every long queue in the building is explained by those three.
  2. 2

    The utilisation term explodes

    This is the part that surprises people, and it is why the equation matters at all.

    • At 50 per cent utilisation the multiplier is 1. At 95 per cent it is 19.
    • At 97 per cent it is 32: queue time is 32 times what it was at 50.
    • Going from 95 to 85 per cent cuts queue time by roughly 70 per cent.

    Ask the room What utilisation are we targeting, and did anyone check what it does to queue?

  3. 3

    The machine you are proudest of

    The one running at 97 per cent is probably the one destroying your schedule.

    • It is generating weeks of lead time that never appear in the utilisation report.
    • They appear in the delivery report instead, where nobody connects them to it.
    • High utilisation and short lead time are mathematically in tension.
  4. 4

    Variability decides what is safe

    You cannot quote a utilisation target without knowing V. The same target means different things.

    • A machine at V = 0.5 can safely run at 90 per cent.
    • The same machine at V = 2.0 has four times the queue at the same utilisation.
    • Aerospace V is structurally high: high mix, variable setups, rework, engineering holds.
  5. 5

    Estimating V without software

    Twenty to thirty observations and a spreadsheet. That is all it takes.

    • Log actual process time for 20 to 30 consecutive jobs, at the machine, not from ERP.
    • Divide the standard deviation by the mean. That is your Ce.
    • Or classify: 0.3 to 0.5 repetitive, 0.5 to 1.5 mixed, 1.5 to 3.0 high-mix with rework.
  6. 6

    Your metrics are fighting your flow

    Every dashboard that rewards utilisation is mathematically in conflict with lead time.

    • OEE targets of 95 per cent, red-amber-green at 90/80/70, shift reports flagging idle time.
    • All of them incentivise the behaviour the equation says will lengthen queues.
    • You either change the metric or you educate the people reading it.

    Ask the room Which of our metrics would punish someone for protecting flow?

  7. 7

    Idle time is the immune system

    Capacity buffer at non-constraints is not waste. It is what absorbs a shock without a queue cascade.

    • When an upstream machine goes down for two hours, buffer capacity elsewhere catches up.
    • Remove all idle time and you remove the ability to recover.
    • Target 85 per cent or below where variability is high. That is the price of flow.

    Ask the room Where in our line is there deliberately spare capacity, and does anyone defend it?