How to identify a bottleneck

This guide helps an SME identify, in two or three weeks of light observation, the point that truly limits the capacity of a flow before spending money on people, machines or software. In Theory of Constraints language, that point is the constraint. In everyday operating language, it is the bottleneck.

What this guide is for

The guide is designed to prevent one of the most expensive mistakes in operations: investing in the point that looks overloaded but does not actually limit the whole flow.

A noisy department, a busy machine or a stressed team may be the visible pain. The bottleneck is the point where throughput is constrained enough to shape the performance of the entire system. These two points do not always coincide.

When to use it

Use it before any capacity investment.

Use it when overall lead time is getting worse while everyone says they are working at full speed. That combination is a classic sign of a misidentified constraint.

Use it when a growth plan requires the company to know what will saturate first as volume increases.

It is also useful before investor discussions. Being able to name the current constraint, the next likely constraint and the cost of elevating them makes an industrial plan more credible.

Before starting

You need at least a rough map of the flow. If you do not have one, start with the guide on how to map a business process.

You also need an explicit agreement with the people involved: the bottleneck is not the culprit. A slow step is often where complexity from the rest of the flow concentrates. If the exercise becomes a search for someone to blame, queues will be hidden and the data will lie.

Say this before starting. It changes the quality of what people show you.

Operational sequence

Step 1 — Follow queues, not visible stress.

For two or three weeks, count the work waiting in front of each step in the flow. Use simple measures: open files by status, unanswered emails by function, jobs waiting in a folder, material on the floor, tickets by owner.

Two observations per week are usually enough, as long as they are taken on the same days and at similar times. The criterion is the queue that grows persistently, not the step where people appear busiest.

Downstream from the constraint, people may work in bursts. Upstream from the constraint, people may be extremely active and still create accumulation. Visible effort and real constraint are different signals.

Step 2 — Separate the queue from the corpse.

Not everything that is stopped is waiting for the next step. Some work is stuck because information is missing, a customer never responded, a decision was suspended, or a file should have been closed but remains open.

Separate real queue from dead work. The first points to the constraint. The second points to entry quality, WIP discipline or unresolved decisions. Mixing them inflates the wrong queue.

Step 3 — Check where accumulation is born.

A queue forms in front of the constraint, but there is a trap: work can accumulate in front of a fast step because the upstream step releases work in waves.

Look at the rhythm of arrival. If the queue grows in irregular bursts, the problem may be upstream: the release pattern. If the queue grows regularly while the step completes less than it receives, the constraint is probably there.

Step 4 — Measure throughput of the candidate.

For the suspected step, count how many units it completes per week and compare that with how many arrive.

If it completes six jobs and receives nine, the constraint is visible and the size of the problem is measurable. The gap between arrival rate and completion rate is the throughput gap. That number will shape the intervention.

Step 5 — Confirm with a temporary elevation test.

Before declaring the bottleneck, test it. Give the candidate step temporary extra capacity: overtime for one week, a borrowed resource, fewer interruptions, or removal of a secondary task.

Then observe the throughput of the whole flow, not only the local step. If the whole flow accelerates, the bottleneck is confirmed. If only the step accelerates and work accumulates immediately after it, the real constraint is downstream. You have avoided a wrong investment at the cost of a small test.

Step 6 — Exploit before investing.

Once the constraint is identified, follow the basic Theory of Constraints sequence.

First, exploit the constraint: remove wasted time, protect it from avoidable interruptions, take away tasks others can do, and keep it supplied with complete input.

Second, subordinate the rest of the flow to its rhythm: stop launching more work than the constraint can absorb.

Only then elevate it with investment. In SMEs the first two moves often recover meaningful capacity at very low cost. Money is the last lever, not the first.

Step 7 — Keep watching the moving constraint.

When one bottleneck is removed or elevated, the constraint moves. That is not bad luck; it is how systems work.

Keep the queue count as a light monthly ritual. The value is not only removing the current bottleneck. It is learning to see the next one before it becomes an emergency.

Expected output

The expected output has three parts.

First, the identified constraint with evidence: stable queue, throughput gap and confirmation test.

Second, a low-cost exploitation intervention with an owner and a date.

Third, a hypothesis about where the constraint will move after the intervention.

If the third part is missing, the exercise is incomplete. Bottleneck work is not one-off improvement. It is a way of reading the system.

Mistakes to avoid

Do not trust visible stress instead of counted queues.

Do not invest before exploiting the constraint.

Do not measure improvement only at the local step. Measure the whole flow.

Do not turn bottleneck diagnosis into a trial of the people working at the constraint.

Do not stop observing after the first success. The constraint will move.

Example

An 18-person printing company believes printing is the bottleneck. Machines are always occupied, and the department is visibly under pressure.

For three weeks the company counts queues. The queue before printing fluctuates but does not grow. The queue before prepress grows from 14 to 23 jobs. One operator checks client files and is interrupted on average 11 times per day by commercial calls asking for status updates.

The confirmation test is simple: for one week all status requests are redirected to a shared report, and the prepress operator is protected from calls. Prepress throughput increases by 25%, and total flow throughput increases by 18%.

No investment was needed. The constraint did not need more capacity; it needed protection. Six months later, after volume grew, the constraint moved to printing. This time the company expected it and bought the second machine based on data, not pressure.

MARTRO connection

In the MARTRO sequence, this guide is the bridge between legibility and intervention. A process map shows the flow. Bottleneck analysis turns the map into an allocation decision: where to put attention and money, and where not to.

In SMEs, the exercise often reveals that the constraint is not a department but a decision point: a signature, an approval, a priority call, or the founder's attention. When that happens, the conversation moves from flow to governance.

Next step

If the constraint is overloaded because too much work is launched upstream, continue with WIP discipline.

If the constraint is a person with unique expertise or signature authority, continue with structured delegation and decision rights.

For the conceptual background, read Bottleneck in the Knowledge library.

Frequently asked questions

How long does it take to identify a bottleneck? Two or three weeks are usually enough for an SME first pass, if observations are consistent and the flow is narrow.

Can the busiest team be a non-bottleneck? Yes. A team can be busy producing work that accumulates elsewhere. Bottleneck diagnosis follows queues and throughput, not perceived effort.

What if there are several queues? There may be several weak points, but one active constraint limits total throughput at a time. Start with the queue whose elevation changes the whole flow.

Should we buy capacity once the bottleneck is found? Not immediately. First exploit and protect the constraint, then subordinate the rest of the flow. Invest only when the constraint still limits the system after those moves.

Why does the bottleneck move after improvement? Because once one point stops limiting the system, another point becomes limiting. That migration is normal and should be monitored.