Answer
How do you do a bottleneck analysis?
Find the station with the longest time per unit once setup is spread over the batch, and check it against where work waits and orders run late: that station is the constraint, and it sets the output of the whole line.
Then try the cheapest remedies at that station first — moving people, shorter repairs, shorter changeovers, less scrap — and buy capacity last. After each change, look again, because the constraint may move.
Reviewed 5 October 2026
How Optonum does it
- The Bottleneck Oracle in the AI Command Center names the constraint as the station with the longest time per unit once setup is spread over its line's lot, checked against late open orders.
- It builds the remedies that apply at that station — moving operators, shorter repairs, shorter setups (SMED), less scrap, the order of campaigns, overtime, another shift, a parallel unit — and tests them on the simulation engine.
- Remedies that cost nothing rank first and investment ranks last. The figures come from formulas and the simulation; the language model only writes the explanation.
- Every proposal waits for a person to approve it. Nothing changes in the plant model until then.
- When the plant is calm, it still names the constraint and the next thing worth measuring.
What it does not do
- A gain the Oracle shows is simulated from the data you entered, not measured on your floor.
- It ranks remedies; it does not prove an optimum.
- It cannot find a constraint the data does not describe: queues exist only where a buffer is drawn in the model.