The production bottleneck — how to find it before you buy a robot

A real-life scenario: the line is missing its targets, so the decision falls on automation. A robot is installed at the station that was easiest to robotise — and after commissioning, the line output does not move by a single percent. The reason is simple: the automated station was not the production bottleneck. The money went where it was convenient, not where the pieces were being lost.

This article shows how to identify and measure a bottleneck without an MES system: which symptoms give it away, how to calculate takt time with a stopwatch and a spreadsheet, and what to expect once the bottleneck is removed. Only with that knowledge does it make sense to talk about the ROI of workstation automation.

What a production bottleneck is and why it sets the line output

The bottleneck is the operation with the longest real takt time in the flow. A line cannot produce faster than its slowest link — no matter how fast the other stations are. If assembly takes 40 seconds per piece and every other operation takes 25 seconds, the line produces one piece every 40 seconds. Full stop.

From this follows the most important investment rule: an hour lost at the bottleneck is an hour lost for the entire factory. A breakdown, a changeover or a break at the bottleneck reduces the day's result one to one. The same events at a station with spare capacity are "made up" by the line without a trace. So before you choose a station to improve, you have to know which one is the bottleneck — otherwise you are optimising something that waits anyway.

Symptoms: how to spot the bottleneck without measurements

Before you reach for the stopwatch, walk along the flow. The bottleneck betrays itself through recurring symptoms:

• a buffer keeps growing in front of the station: pallets, containers, trolleys full of parts queuing up,
• the stations behind the bottleneck wait — operators "look for something to do", machines stand ready without material,
• overtime is localised: one crew or one station stays late, not the whole shop floor,
• the plan is rescued with "due yesterday" batches pushed out of sequence through this one station,
• the shift leader watches this station personally, because "when it stops, everything stops".

One symptom is a clue, several at once are near certainty. But note: observation tells you where to look — it does not yet tell you how much is missing and why.

How to locate the bottleneck without an MES system

For a reliable diagnosis you do not need MES or IIoT sensors. A stopwatch, a spreadsheet and two or three shifts of observation are enough.

1. Sketch the flow: all operations in order, from raw material to dispatch, including transport and inspection.
2. Measure the cycle time of each operation over 10–20 consecutive pieces — not just one, because a single measurement always lies.
3. Record stoppages separately: material shortages, changeovers, breakdowns, waiting for an operator, rework.
4. Calculate the real takt time of the station: the available shift time divided by the number of good pieces from that shift.
5. Compare the takt times — the station with the longest real takt time is your bottleneck.

The key word is "real". For example: a welder with a 20-second cycle that stands still for 45 minutes twice per shift during changeovers and loses pieces to rework can have a worse real takt time than the manual assembly next to it. The machine's catalogue output does not matter — what counts are the good pieces at the end of the shift.

A simple numerical example: four operations with cycle times of 25, 32, 40 and 28 seconds. In theory, the third operation is the bottleneck. But if the second operation loses an hour and a half to changeovers in an eight-hour shift, its real takt time rises, indicatively, to close to 40 seconds and draws level with the third operation. The conclusion from measurement can therefore differ from the conclusion from observation alone — which is why you measure all the operations, not just the suspect.

Symptom, cause, measurement — a diagnostic cheat sheet

The last row often comes as a surprise: sometimes the bottleneck is no machine at all, but moving parts between stations by trolley. Then the answer is not a robot but conveyor systems and buffering — more on selection in the article how to choose a conveyor for your process.

The typical mistake: automating a station that is not the bottleneck

The most common misdirected-investment scenario looks like this: the operation chosen for automation is the most strenuous one, the one complained about the loudest, or the simplest to robotise. The robot packs parts faster than a person — but packing had spare capacity, so the robot simply waits longer. Line output unchanged, cost incurred, and the argument "automation didn't work for us" sticks around for years.

The correct order is the reverse: first measure the takt time of all operations, then point to the bottleneck, and only then ask whether a robot is the best solution at the bottleneck — or maybe a shorter changeover, a second fixture, a buffer or a change in organisation. Automation is an answer to a measured problem, not to an impression. How to prepare the process data before talking to an integrator is described in the article how to prepare a process for automation.

At the bottleneck itself, it also pays to exhaust the cheaper options first: shortening the changeover, a parallel fixture, moving some tasks upstream or downstream of the bottleneck, an extra pair of hands at peak times, a buffer decoupling it from the neighbouring stations. Only when these moves are not enough or do not last does investing in automating the bottleneck have a strong justification — and then it usually holds up in the payback calculation as well.

The bottleneck moves — and that is normal

Removing a bottleneck does not close the topic. When assembly speeds up from 40 to 25 seconds, the next slowest operation becomes the bottleneck — for example, inspection with a takt time of 30 seconds. The line speeds up, but only to the new limit.

That is why, before investing, it is worth calculating two steps ahead: what takt time the line will have after the improvement and whether that is enough for the demand. Indicatively, if the bottleneck runs at 40 s and the next operations at 30 s, speeding the bottleneck up below 30 s gains nothing more — the surplus capacity turns into waiting. This simple calculation often changes the scope of the project: instead of one expensive robot, two smaller changes at two stations work out better. The same analysis also tells you whether the new line takt will withstand seasonal order peaks or only the average load — a frequent cause of disappointment six months after implementation. A good automation partner starts precisely with this kind of flow analysis, not with selecting a robot.

Summary

You can find a production bottleneck without MES: first the symptoms (a buffer in front of the station, waiting machines behind it, localised overtime), then a stopwatch measurement of the real takt time over a dozen or so pieces, and finally a comparison of the operations. Automate only what has been measured — and calculate where the bottleneck will move after the change.

Want to consult on which station really limits your line? Describe the process via the contact form — we will respond with a proposed approach and a quote within 48 hours.