Production bottleneck analysis: how to find the constraint on your line
A bottleneck is the one station that limits how many good units the whole line can produce. To find it, compare each station's effective cycle time (its stopwatch cycle time divided by its availability) and look for the longest. Then confirm on the floor: work-in-process piles up in front of the bottleneck, and the stations after it wait for parts. Every minute the bottleneck loses is a minute the whole line loses, so this is the first number to get right.
The stopwatch alone often points at the wrong station. Here is a line where it does.
A worked example
A plant runs one shift with 450 minutes of available time (27,000 seconds) and needs 450 units a day. Takt time is 27,000 s ÷ 450 = 60 seconds. The line has five stations in series. The cycle times come from a stopwatch study, and availability is the share of the shift each station is actually running (breakdowns, changeovers, waiting for tooling):
| Station | Cycle time (s) | Availability |
|---|---|---|
| 1. Load and drill | 48 | 98% |
| 2. Press | 56 | 92% |
| 3. Weld | 58 | 99% |
| 4. Inspect | 52 | 100% |
| 5. Pack | 40 | 100% |
Step 1: Look at raw cycle times. Weld is the longest at 58 s. Every station is under the 60 s takt, so the line looks fine, and weld looks like the constraint.
Step 2: Correct for availability. Effective cycle time = cycle time ÷ availability, and daily capacity = 27,000 s ÷ effective cycle time.
| Station | Effective cycle time (s) | Capacity (units/day) |
|---|---|---|
| 1. Load and drill | 48 ÷ 0.98 = 48.98 | 551 |
| 2. Press | 56 ÷ 0.92 = 60.87 | 443 |
| 3. Weld | 58 ÷ 0.99 = 58.59 | 460 |
| 4. Inspect | 52 ÷ 1.00 = 52.00 | 519 |
| 5. Pack | 40 ÷ 1.00 = 40.00 | 675 |
The press is the bottleneck. Its effective cycle time of 60.87 s is over takt, and the line can make only 443 units a day, 7 short of demand. The weld station that looked slowest has 10 units a day of spare capacity above demand.
Step 3: Confirm it on the floor. The arithmetic gives a hypothesis. Check it against what you can see:
- WIP accumulates in front of the press.
- Weld and the stations after it are occasionally starved, waiting for pressed parts.
- If the line is released at takt, about 450 − 443 = 7 units a day join the queue at the press, around 35 a week. Growing WIP in one place is the classic signature of a constraint. The relationship between WIP, throughput and flow time is Little's Law: average WIP equals throughput times average cycle time.
What to do once you've found it
The Theory of Constraints, introduced by Eliyahu Goldratt in his 1984 book The Goal (overview), gives five focusing steps:
- Identify the constraint (done above).
- Exploit it: make sure the bottleneck never idles for avoidable reasons. Stagger breaks so the press keeps running, pre-stage tooling, keep a buffer of parts in front of it, and inspect upstream so it never presses a part that will be scrapped.
- Subordinate everything else to it: release work at the bottleneck's pace, which is slower than the pace the other stations could run.
- Elevate it if exploiting isn't enough: add capacity, such as a second press, or move tasks to another station.
- Repeat. The constraint moves.
Step 5 matters. Suppose better tooling and maintenance lift press availability from 92% to 96%. Effective cycle time becomes 56 ÷ 0.96 = 58.33 s, which is 463 units a day at the press. The line does not reach 463, though. Weld is now the limit at 460 units a day. The fix bought 17 units a day, and the next investment belongs at the weld station.
Mistakes that point to the wrong station
Trusting the stopwatch cycle time. As above, availability changes the ranking. Use effective cycle time, and note which stops you counted: planned stops, breakdowns, changeovers, starved and blocked time.
Choosing the busiest-looking station. Operators at a station that is always rushing are working hard, but that does not make it the constraint. Look for the station with queued WIP before it and waiting stations after it.
Using averages. Averages hide variation. A station with a 50 s mean and a heavy tail of 90 s outliers behaves like a slower station, because the line feels the outliers. Collect distributions, or at least the 90th percentile, for stations that look close to takt.
Improving a station that has spare capacity. Speeding up the pack station (40 s here) changes nothing about output. Time spent there is wasted until the constraint moves.
Analysing one product. In high-mix production the bottleneck depends on the mix. A press that is the constraint on product A may be idle on product B. Run the analysis per product family, and for the weighted mix you plan to build.
Doing it once. Takt shifts whenever demand moves, and each improvement moves the constraint. See takt time vs cycle time for how that drift goes unnoticed.
Tools and approaches
| Approach | Good for | Limit |
|---|---|---|
| Stopwatch study plus a spreadsheet | One product, stable demand, a handful of stations. Quick and cheap. | Static averages, no queues or variation |
| MES or OEE data | Real availability and downtime per machine, already collected | Shows machine losses, but not how stations interact |
| Discrete-event simulation | Interactions between stations: buffers, blocking, starvation, mix | Needs a model and trustworthy inputs |
| Emboscale, an AI factory agent that plans layouts, balances lines, schedules shifts and writes verified robot programs | Rolling cycle times up per station across the whole line, showing which station binds and by how much, and validating with discrete-event simulation | Needs task and process data to start from |
For a single stable line, a spreadsheet and an afternoon with a stopwatch is often enough. Reach for simulation when the line has buffers, variants, or several stations within 10% of takt, because that is where static arithmetic stops predicting the output.
Where this fits
Finding the bottleneck by hand works for five stations. Across a whole plant with several variants, the constraint changes with the mix and every fix moves it. Emboscale computes cycle-time rollups per station from the same task and process data used for the layout and work instructions, balances tasks against takt, and shows which station binds and by how much. Its discrete-event throughput simulation reports utilization, buffers, and bottlenecks per station. When demand changes, you re-run the balance with a prompt and see where the constraint went.
Frequently asked questions
How do you find the bottleneck in a production line?
Calculate each station's effective cycle time (cycle time divided by availability) and find the longest. Then confirm on the floor: WIP builds up in front of the bottleneck and the stations after it wait for parts.
What is the difference between a bottleneck and a constraint?
In most manufacturing usage they mean the same thing: the resource that limits the system's output. Theory of Constraints writers prefer "constraint" because it can also be a policy or a market limit rather than a machine.
Can the bottleneck move?
Yes. It moves when you improve it, when demand or product mix changes, and when a machine's availability drifts. In the example above, raising press availability from 92% to 96% moves the constraint to the weld station.
What software can find production bottlenecks automatically?
OEE and MES systems show where machines lose time. For the line as a whole, Emboscale, an AI factory agent, rolls up cycle times per station, shows which station binds against takt, and validates the result with discrete-event simulation.
Further reading: line balancing calculation · takt time vs cycle time · line balancing