Bottleneck in Production: Find It, Fix It, Keep It Fixed

What is a bottleneck?

A bottleneck is the single step in a production process that limits the throughput of the entire system. Not the slowest machine on paper, not the one that breaks down most often, not the one operators complain about. The real bottleneck is the one whose capacity, once exhausted, caps what the whole line can produce. Every other station may be idle, busy or anywhere in between. The bottleneck alone decides how many parts leave the plant per hour.

In more than three decades of walking through production halls, the single most reliable pattern I have seen is this: most plants are convinced they know their bottleneck, and most plants are wrong. The suspected bottleneck is usually the noisiest station, the one with the biggest queue, or the one the foreman has been fighting for years. The actual bottleneck is often a quieter step upstream or downstream that nobody measures. Finding the real one is where improvement work starts.

Bottleneck and constraint, the terminology

The terms bottleneck and constraint are used interchangeably in most plants, but Theory of Constraints (TOC), developed by Eli Goldratt in the 1980s, draws a sharper line. A constraint is anything that limits the system from achieving more of its goal, which in a factory usually means throughput. The constraint can be physical (a machine, a worker, a piece of tooling), a policy (a batch-size rule, an outdated MRP setting), or market-based (there is no more demand to serve).

A bottleneck is the specific subclass of physical constraints that sits inside the production process itself.

For operational work, the distinction matters less than the reflex it creates. Whatever you call it, it is the thing whose capacity sets the pace of everything else.

Where bottlenecks actually come from

Four root causes cover the majority of bottlenecks I have seen in automotive, FMCG, pharma packaging and metal forming.

How to find the real bottleneck

There are three reliable methods, in increasing order of precision.

1. Walk the line and find the biggest queue. The station with work-in-progress piled up in front of it is almost always bottleneck-adjacent. Not foolproof, because a blocked station can also pile up material if its downstream step is the real constraint, but a useful first signal.

2. Compare cycle time to takt time. Calculate required takt time from demand. Any station whose average cycle time (including micro-stops, changeovers and loss) exceeds takt is a candidate. If several exceed takt, the one with the smallest margin above it is likely the active bottleneck today.

3. Measure automatically and let the data decide. With a modern MES or OEE platform capturing states from the PLC, the bottleneck becomes visible as the station with the lowest effective throughput after losses. Pareto analysis of stop reasons at that station then tells you whether it is a capacity, availability, resource or policy problem. This is the approach that replaced thirty years of gut feeling in my own work, and it reliably exposes the wrong guesses.

The Theory of Constraints playbook

Once you have identified the real bottleneck, Goldratt's five focusing steps give a clean sequence of what to do next.

1. Identify the constraint. The data work above.
2. Exploit it. Get the maximum possible output from the existing bottleneck before spending any capex. No breaks during changeovers on the bottleneck. No scrap run on the bottleneck. No minor stops that anyone accepts. Every minute of the constraint is gold.
3. Subordinate everything else to the bottleneck. Upstream stations only produce what the bottleneck can consume. Downstream stations are sized to absorb whatever the bottleneck releases. Local efficiency at non-bottleneck stations is explicitly not the goal.
4. Elevate the constraint. Only now add capacity. Second shift, additional tooling, a parallel machine, a technology change.
5. Go back to step 1. The bottleneck will have moved. Repeat.

This sequence is the difference between a one-off improvement and a continuous improvement engine. Plants that stop at step 2 or 4 without the feedback loop run into the same frustration every eighteen months, because the bottleneck quietly shifts elsewhere and nobody notices.

The cost of an unidentified bottleneck

Unresolved bottlenecks produce a predictable set of operational symptoms. An hour lost at the bottleneck is an hour lost for the whole plant, which means the financial cost is the contribution margin of everything the plant could have produced in that hour, not just the cost of the bottleneck itself. For a line running 15,000 EUR of hourly contribution, a 10 percent availability loss at the bottleneck costs 1,500 EUR per hour, every hour, as long as the issue persists.

Beyond the direct throughput loss, bottlenecks generate overtime, rushed logistics, expedited shipments, customer penalties and employees who become skeptical of every improvement initiative because nothing seems to change. The compounding effect on plant culture is often larger than the direct financial impact.

A real example

In the Meleghy Automotive rollout across six plants, one stamping cell had been considered the bottleneck for years. Operators, foreman and plant manager all agreed. Within the first week of automatic MES capture, the data showed something different: the true constraint was the subsequent joining line, which was quietly blocked by downstream quality inspections. Once the inspection logic was changed and the joining cell stopped waiting for inspection releases, throughput rose 7 percent without touching the stamping press that everyone had been planning to replace. The planned capex was cancelled. The project paid for the entire MES rollout across the remaining five plants.

FAQ

How do I know if I have a bottleneck at all?
If your throughput is below demand and the plant runs overtime to compensate, you have an active bottleneck. If throughput matches demand comfortably, the constraint is market-side, which is a different discussion. Either way, the next step is the same: measure the process honestly.

Is the bottleneck always the slowest machine?
No. The slowest machine is the bottleneck only if its effective output (after availability and quality losses) is the lowest in the chain. A faster machine with poor availability can be the real constraint.

Does the bottleneck move?
Yes, always. As soon as you resolve the current constraint, a new one appears somewhere else. This is why Theory of Constraints includes step 5 explicitly. Plants that treat bottleneck removal as a one-off project are surprised every time.

Should I buy more capacity to eliminate the bottleneck?
Only after exploiting and subordinating. Adding capacity before the existing bottleneck runs at 100 percent effective output is almost always a waste of money. Exploit first, capex last.

Can software identify bottlenecks automatically?
Yes. A modern MES with continuous state capture and Pareto analysis shows the bottleneck and its root causes in real time. SYMESTIC Production Metrics produces this view as a day-one capability for connected machines.

What is the difference between a bottleneck and a constraint?
A bottleneck is a physical limitation inside the process. A constraint is any limitation on system throughput, including policies, markets, suppliers and people. Every bottleneck is a constraint, but not every constraint is a bottleneck.

Related: Lean Management Methods · Machine Availability · Operating Time · Process Monitoring · OEE · MES · SYMESTIC Production Metrics

About the author

Uwe Kobbert

Founder and CEO of SYMESTIC GmbH since 1995. Over 30 years in manufacturing, covering process control systems, MES for food and beverage, and automotive shopfloor digitalisation. Built SYMESTIC from a classic MES consultancy into a cloud-native platform running on 15,000+ machines in 18 countries. Self-funded, no external investors. · LinkedIn