I once looked at a furniture factory making upholstered sofas.
The bottleneck, everyone agreed, was the carpentry department, which built the wooden frames. It was always busy. Orders stacked up. Deliveries slipped.
When pressure built, they did what many factories do. They invested in a robotic workstation to automate frame production.
Expensive. Technically impressive. Throughput barely moved.
The robot was not slow. The problem was never speed.
The factory produced dozens of sofa models, each with a different frame geometry. That meant frequent changeovers, complex setups, constant adjustments.
The robot was fast when doing the same thing repeatedly. The system never gave it that stability.
So instead of increasing output, the investment:
- Raised switching costs
- Locked flexibility into hardware
- Made variation more painful, not less
Carpentry looked like the bottleneck. In reality, the system was choking on variety upstream, long before work ever hit the department.
Nothing was wrong with the department. The system asked it to do too many different things at once.
The most convincing bottlenecks are often the wrong ones. And once you build automation around a false assumption, it becomes very expensive to admit it.