How to Find a Business Process Bottleneck From Your Data
Key takeaways
- Record capacity, actual throughput, demand, cycle time, and WIP per stage over one consistent period before calling any stage a bottleneck.
- A true constraint shows high utilisation, growing WIP in front of it, and throughput below upstream stages; a busy stage that clears its queue is not constraining anything.
- Quantify a constraint's cost in units per period as system capacity minus bottleneck capacity, then convert backlog growth to lead time using Little's Law.
- Run the fix sequence in order: offload first, fix variation or structural loss second, add capacity last, and only after pricing the gap in units.
- Verify a fix by re-measuring end-to-end throughput with the same protocol, not utilisation at the stage you changed, then re-run the template to find where the constraint moved.
You can locate a business process bottleneck from one week of real data: record each stage's capacity and demand per period, compute utilisation, and the stage running nearest 100% with WIP piling behind it is your constraint. This walkthrough gives you the fill-in template, a worked numeric example, and a throughput-based rule for fixing, offloading, or elevating it.
What data do you need before you can call a stage a bottleneck?
Five numbers per stage, over one consistent period: capacity, actual throughput, demand, cycle time, and WIP. Anything less and you are guessing, which is exactly the trap in the forum thread where a requester was told "I don't believe you have presented enough information to make even a guess."
Capacity vs actual throughput per stage
Capacity is what a stage could produce if fully loaded; actual throughput is what it produced. A stage running at 95% utilisation while downstream stages sit at 60% is your constraint candidate. Note the distinction matters: a bottleneck is the busiest stage, but a constraint is the stage that limits total system throughput, per standard Theory of Constraints definitions.
Where to pull the numbers without new software
Export ticket or job timestamps from your existing tools: Jira, Trello, or a job sheet gives cycle time and throughput; timesheets or machine logs give capacity; a weekly count of open jobs gives WIP and backlog. For the template that structures these numbers into a constraint diagnosis, see Improve First's bottleneck guide.
How do you distinguish a bottleneck from a busy-but-adequate stage?
A stage that runs at high utilisation is not automatically your business process bottleneck. Three conditions must hold together: utilisation is high relative to the demand gap, WIP is accumulating in front of the stage, and its throughput sits below the stages upstream of it. Any one alone can mislead you.
The three-part test
| Signal | Busy-but-adequate stage | True constraint |
|---|---|---|
| Utilisation | High, but demand gap is closed | High, demand gap open |
| WIP in queue | Cleared each cycle | Grows shift on shift |
| Throughput vs upstream | Matches or exceeds | Lower than upstream |
Worked example: two busy stages, one constraint
Say Review runs at 95% utilisation and Approval runs at 92%. Review clears its queue every shift, so WIP never builds. Approval's queue grows by four items per day while upstream stages push through twelve. Approval is the constraint; Review is just heavily loaded. Run both stages through the three-part test on the Constraint Impact Calculator before quoting a fix, and you avoid spending the client's money on the wrong stage.
How do you quantify the cost of a bottleneck in units of throughput?
The cost of a constraint is the gap between what the system could ship if the constraint ran flat out and what it actually ships. Express it in units, not currency: units per day, cases per week, tickets per sprint. Currency invites argument about margins and labour rates; units keep the scoping conversation on the process.
The fill-in throughput template
Fill in four lines from your stage-level data:
- System capacity (demand or demonstrated upstream capacity): ___ units/period
- Bottleneck capacity: ___ units/period
- Lost throughput = System capacity − Bottleneck capacity = ___ units/period
- Backlog change = Lost throughput × number of periods = ___ units
Then convert to lead time using Little's Law: average lead time equals average WIP divided by average throughput (Little's Law, MIT Sloan teaching note). If the backlog grows, WIP grows, and lead time stretches with it.
Worked numbers: a 12-unit/day constraint in a 20-unit/day process
A five-stage workflow ships 12 units/day because stage three caps at 12, while demand sits at 20. Lost throughput is 8 units/day. After ten days the backlog is 80 units; at a steady WIP of 40 units, lead time has moved from 2 days toward 6.7 days and climbing. That arithmetic, run in minutes on Improve First's calculator, is usually enough to scope whether the constraint is worth a deep engagement.
When should you fix a bottleneck versus offload work around it?
Run three tests in order: offload, fix, elevate. Offloading is cheapest, fixing is next, adding capacity is last. Skipping the sequence is how consultants quote a six-week engagement for a problem a two-day triage would have scoped correctly.
Offload test: transferability and headroom
Ask two questions. Can the constraint task be done by another stage or person without new skills or equipment? And does the receiving stage have headroom, meaning utilisation below roughly 85%? Both yes: offload. One yes: you have found the shape of your fix, not the fix itself.
Fix test: is the loss structural or variation-driven?
If the constraint stage loses capacity to rework, changeovers, or absence, that is variation, and standard work or batch-size changes recover it. If it runs at full utilisation and still cannot meet takt time, the loss is structural. Structural constraints need a redesign, not discipline. The Constraint Impact Calculator separates the two by comparing stage cycle time against demand gap.
When elevation is the only remaining option
If demand gap persists after offloading and fixing, add capacity: a second machine, an extra pair of hands, extended hours. Quantify the gap in units per week first, then price the elevation against that number. Anything else is guesswork.
How do you verify a bottleneck fix actually raised end-to-end throughput?
Measure the system, not the stage
Re-measure using the same protocol you used to diagnose: same period length, same demand mix, same definition of throughput. Compare finished units per week, not utilisation at the stage you fixed. A stage can drop from 98% to 80% utilisation while weekly output stays flat, because the work simply queued somewhere else. Queueing theory explains why: as utilisation approaches capacity, wait times grow non-linearly, so the constraint stage will always look "busy" even when it is no longer limiting output (Leyland Pitt and colleagues' summary of utilisation thresholds). Judge the fix on system throughput against the demand gap you calculated earlier.
Re-run the template: where the constraint moves next
Fill the same template with post-change data. If total throughput rose, the constraint has moved; the stage now showing the largest gap between capacity and demand is your next candidate. Note it in the client-ready scoping report, because it sets the scope of the next engagement. The worked example at https://improvefirst.com/examples/invoice-processing-capacity shows this re-run in practice.
What recurring signals indicate a bottleneck will re-form after you clear it?
Demand-side signals
Seasonal demand peaks and batch-release cadence are the two demand patterns that most often rebuild a constraint you just cleared. If throughput at the stage spikes above its measured capacity for a week or more each cycle, log the peak-to-capacity ratio in your diagnostic notes. Batch releases upstream create the same effect in miniature: a stage idles, then absorbs a queue. Record release frequency and batch size alongside cycle time so the scoping report shows when the constraint returns, not just whether it exists.
Resource-side signals
Single-skilled operators are the classic re-form trigger: the constraint moves to whoever holds the only copy of a skill the moment volume shifts. Note skill coverage per stage during data collection. Upstream improvements also push more WIP downstream, so re-measure the next stage after any fix. The Calculator's constraint-movement preview flags where throughput will pool next, which is the question the CR4 thread above never got answered.
Which tools support this diagnostic, and where does each fall short?
Running the same criteria across tools
A spreadsheet capacity model handles the core diagnostic: throughput and cycle time per stage, demand gap per stage, and a simple sum to locate the constraint. It cannot show where the constraint moves after a fix, and every scenario means rebuilding formulas by hand. Improve First's Constraint Impact Calculator covers the same steps plus constraint-movement projection and a client-ready scoping report, but it works from stage-level averages, not distributions, so it will miss variability-driven constraints. ProcessModel simulates variability and interdependencies, which suits deeper engagements, though it costs more and takes longer to learn than a scoping call justifies. DO.Impact and FlowVisual both emphasise visualising process flow; neither, per their vendor pages, quantifies constraint movement in units of throughput the way the diagnostic above requires. KaiNexus tracks improvement work after scoping, not the diagnosis itself.
Where Improve First's calculator falls short
It gives a static picture: no queues, no arrival variation, no WIP aging. If a client's constraint hides in variation rather than capacity, the calculator will point at the wrong stage, and only simulation or direct observation will catch it.
How does the fix-first approach change the order of steps you tackle?
Traditional scoping runs proposal first, diagnosis second. Fix-first inverts that: measure throughput and cycle time per stage, locate the constraint, clear the cheapest one, re-measure, and only then talk about a deeper engagement.
The cheapest constraint is rarely the most visible one. A stage running 20% under takt time because of one handoff delay may cost less to fix than a capacity shortfall that needs new equipment.
Re-measure before you quote anything. If end-to-end throughput did not move, the constraint moved, and your next recommendation changes.
Scoping the engagement from measured impact
A proposal backed by numbers the client can check, in units of throughput they already track, sells itself. Where Improve First's Constraint Impact Calculator falls short: it locates and sizes the constraint but will not simulate a fix for you, so quote deeper work only after the re-measurement proves the system moved.
FAQ
Can I find a business process bottleneck with just a spreadsheet?
Yes. A spreadsheet handles the core diagnostic: record capacity, throughput, demand, cycle time, and WIP per stage, then compute utilisation and the demand gap. What it cannot do is project where the constraint moves after a fix, and every new scenario means rebuilding formulas by hand.
How long should I collect data before diagnosing a bottleneck?
One week of real data is enough, provided you capture all five numbers per stage over a consistent period: capacity, actual throughput, demand, cycle time, and WIP. Shorter than that and you risk mistaking a demand spike or batch release for a structural constraint.
What is the difference between a bottleneck and a constraint?
A bottleneck is simply the busiest stage. A constraint is the stage that limits total system throughput. They are often the same stage, but not always: a heavily loaded stage that clears its queue every shift is a bottleneck without constraining anything. The three-part test separates them.
How do I calculate bottleneck capacity utilisation?
Divide the stage's actual throughput by its capacity for the same period, then multiply by 100. A stage at 95% utilisation while downstream stages sit at 60% is your constraint candidate. Confirm it with the other two signals: WIP growing in front of it and throughput below upstream stages.
Do I need simulation software to find a bottleneck?
No. Stage-level averages from a spreadsheet or calculator locate and size most constraints. Simulation earns its cost only when the constraint hides in variation rather than capacity: queues, arrival spikes, or WIP aging that averages miss. If utilisation and demand gap look clean but output stays flat, reach for a simulator.