What is moving?
Material, requests, approvals, attention?
Bottlenecks require a defined flow and desired throughput.
Side 80 · Special
Bottlenecks
A study of constraints that govern system performance. Bottlenecks appear when flow encounters limited capacity, causing queues, delay, underused downstream resources and misleading incentives to optimize everything except the true constraint.
flow→capacity→constraint→queue→system effect
06bottleneck lenses
05constraint types
05improvement traps
80Side
The system cannot sustainably outperform its active constraint.
Local capacity elsewhere may create idle time or inventory without increasing final output.
How much can each stage process?
Per unit time.
Nominal and effective capacity may differ because of variability and downtime.
Which stage limits completion?
System, not local, view.
The active bottleneck is the resource whose capacity binds system output.
Where does unfinished work accumulate?
Visible symptom.
Persistent upstream queues often reveal a constrained stage.
What reaches completion?
End-to-end output.
System throughput, not utilization everywhere, is the primary performance measure.
Queues are the shadow cast by constrained capacity and variability.
Even when average capacity exceeds average demand, variability can create significant waiting.
How quickly work enters.
Demand bursts can overload a stage temporarily.
How quickly work exits.
Slow or variable processing increases queue formation.
How close to full capacity?
Waiting often rises sharply as utilization approaches 100%.
How uneven are arrival and service times?
More variability increases waiting at the same averages.
Where does waiting reside?
Systems trade inventory, time and spare capacity differently.
Who gets served first?
Queue discipline changes delay distribution even when total capacity is unchanged.
Constraints are not always machines.
The binding limit can be physical, informational, cognitive, financial or institutional.
Physical
Machine, lane, server, room.
Hard capacity limits create visible saturation.
Human
Specialist skill or attention.
Rare expertise often becomes the true capacity constraint.
Information
Missing or delayed knowledge.
Work can stall even when physical capacity is abundant.
Policy
Approval or rule constraint.
Governance can deliberately limit throughput for quality, safety or control.
Market
Demand itself can be the constraint.
Increasing production capacity has no value when customer demand is binding.
Bottlenecks recur across domains because capacity is always finite somewhere.
The vocabulary changes, but the system logic is often similar.
| Domain | Flow | Possible bottleneck | Symptom |
|---|---|---|---|
| Manufacturing | Parts | Machine / setup | WIP accumulation |
| Computer systems | Requests / data | CPU, disk, network, lock | Latency spike |
| Organizations | Decisions | Approver / specialist | Backlog |
| Transport | Vehicles / passengers | Lane, station, junction | Queue / congestion |
| Biology | Metabolic flux | Enzyme / substrate availability | Accumulating intermediate |
| Learning | Information → usable knowledge | Attention / working memory | Overload |
Removing one bottleneck usually reveals another.
Constraint removal changes the system rather than ending the existence of constraints.
Add capacity at the constraint.
The original queue shrinks until another resource binds.
Speed one stage.
Automation may simply move waiting to the next manual or approval stage.
Add simultaneous processing.
Shared upstream or downstream resources may become new limits.
Reduce variation.
Lower variability can increase usable capacity without adding resources.
Reshape arrival pattern.
Appointments, pricing or batching can flatten peaks.
Remove the constrained step entirely.
Architectural change can dominate incremental optimization.
Improvement should follow the constraint, not local utilization.
Optimizing a non-bottleneck can increase work-in-process while leaving final throughput unchanged.
Identify
Find the stage that truly limits end-to-end performance.
Exploit
Protect the constraint from avoidable downtime and low-value work.
Subordinate
Align upstream and downstream behavior to the constraint’s pace.
Elevate
Add capacity only after existing constraint capacity is used well.
Repeat
Find the new constraint after the system changes.
The GoalGoldratt & Cox · theory of constraints
Factory Physicsflow, variability and bottlenecks
Queueing Theorycapacity and delay
Operations Managementconstraint and throughput analysis