Bottleneck Analysis: Find the Constraint That Limits Flow
A bottleneck is the step that currently limits system throughput, not simply the place with the largest visible queue.
How to identify a bottleneck
Define the unit of flow and system boundary, then measure arrival rate, completion rate, work in progress, waiting time, and effective capacity at each step. Confirm the constraint by changing or protecting its capacity and observing the whole-system result. After improvement, measure again because the bottleneck can move.
Find the step that limits the whole flow. Improve that step first, then measure again because the bottleneck may move.
Key relationships
- A bottleneck limits the output of the full process.
- Work in progress builds before a step that cannot keep up.
- A bottleneck change moves the active constraint to another part of the flow.
Choose the flow unit and outcome
A process may move requests, patients, batches, decisions, or data records. Pick one unit and define when it enters and leaves the system. Mixing cases with different work requirements can make a fast step appear slow or conceal a difficult case mix.
Use an outcome that matters to the full flow, such as completed correct cases per day, not activity within one department. Local output that increases downstream rework is not additional throughput.
Measure demand, capacity, and variability
Arrival rate describes demand. Effective capacity is the sustainable completion rate after normal interruptions, changeovers, errors, and availability. Averages alone hide bursts, so inspect time patterns and the spread of work duration.
A step operating near its effective limit can create rapidly growing waits when demand varies. The queue may appear downstream because of batching, even when the capacity constraint is earlier.
Distinguish constraint signals from symptoms
Signals include persistent work waiting before a step, high utilization, downstream starvation, and system output that changes when the step changes. A large queue can also result from release policy, priority rules, missing information, or work that should never have entered the system.
Observe where items spend time and why. Classify waiting as capacity, dependency, batching, rework, approval, or scheduling. Different waits require different interventions.
Exploit and protect the current constraint
Before adding capacity, reduce avoidable loss at the constraint. Ensure inputs are complete, keep specialized work away from noncritical tasks, schedule maintenance deliberately, and prevent defective work from consuming scarce time.
Control release into the system so unfinished work does not overwhelm the constraint. A smaller, visible queue with clear priorities can improve completion even when nominal staffing stays the same.
Verify the system effect and repeat
Run a limited change and track total completions, end-to-end time, quality, and work in progress. If the local step improves but system output does not, another constraint or policy is limiting flow.
When throughput rises, the constraint may move. Recreate the map rather than continuing to optimize the old bottleneck. Improvement is a cycle of identify, protect, elevate, and measure again.
Separate touch time from elapsed time
Touch time is active work on an item. Elapsed time includes waiting, batching, rework, and handoffs. A step with ten minutes of touch time can create a four-day delay when work waits for a weekly batch. Measuring both values prevents a labor estimate from being mistaken for a flow diagnosis.
Sample individual item journeys as well as step totals. Averages can combine fast routine items with rare complex items and hide the class that occupies the constraint. Segment only when the categories represent a real difference in required work or priority.
Read flow data without mistaking every queue for a bottleneck
Consider a three-step request process measured over the same working period. Intake completes 60 cases, specialist review completes 24, and final approval completes 23. A visible approval queue may still be a symptom rather than the active constraint.
| Step | Completed cases | Active work time per case | Observed waiting |
|---|---|---|---|
| Intake | 60 | 6 minutes | Little waiting before intake |
| Specialist review | 24 | 25 minutes | Ageing queue before review |
| Final approval | 23 | 8 minutes | Small batch waits after review |
Specialist review is the leading constraint candidate because work accumulates before it and downstream output closely follows its rate. The approval wait may come from batching, but approval has enough observed capacity to keep pace. Confirm the diagnosis by protecting review time or improving input quality, then measure total correct completions and end-to-end time.
Reject a local improvement when
- The step works faster but total completed output does not rise.
- End-to-end time stays flat because waiting moves elsewhere.
- Rework or error rates increase.
- Work in progress grows even though local utilization looks better.
Check that flow is the real problem
Use the problem-solving method selector when you are unsure whether the task needs flow analysis, causal diagnosis, or option comparison.
The meeting room bottleneck challenge provides a compact case for identifying the active constraint before proposing capacity.