Poka-Yoke: the mistake-proofing system, explained
Shingo started from an idea that makes a lot of managers uncomfortable: your operator is going to make a mistake. So he designed the process so he couldn't.
A poka-yoke is a device or a design that makes it impossible to get it wrong. Or that leaves the mistake in plain sight the instant it happens.
Enter your email and you will get an alert when something new is published. Unsubscribe anytime.
It comes from the Japanese poka (unintentional mistake) and yoke (avoid). The engineer Shigeo Shingo developed it at Toyota from the sixties on.
And his starting point is what separates it from any ordinary quality plan: it takes for granted that the person will make a mistake. Not through poor training or carelessness. Because across eight hours of repetitive work, a mistake is statistically inevitable.
If that's a fact, then blaming the operator fixes nothing. What has to be redesigned is the process.
The three possible moments
| Approach | When it acts | Cost of the fault |
|---|---|---|
| Reactive | when the defect already reached the customer | maximum: complaint, recall, reputation |
| Proactive | at the later inspection | medium: rework or scrap |
| Preventive (poka-yoke) | before the error happens | practically nil |
Inspection, however good yours is, finds the defect after you've made it. The poka-yoke acts before. And you know how this goes: the later you catch it, the more it costs.
The two types
Control — the process physically won't let you get it wrong. The part only fits one way round. The connector only goes in the right way up. The machine won't start if the bolt is missing. This is the good one: it doesn't depend on anybody reacting in time.
Warning — the error is possible, but the system sings out immediately with a light, a beep or a lock. You use it when preventing it physically isn't worth the money.
A good poka-yoke is usually cheap and obvious: a guide, a stop, a counter, a change of shape. And here's a practical rule for you: if your solution needs a new procedure and a training course, it isn't a poka-yoke. It's one more instruction nobody reads.
The catch?
That for sixty years it's had a very clear limit: it only works on errors you can foresee and model physically.
You can design a fixture that stops the part being mounted backwards. You cannot design a stop that prevents a seal arriving with a micro-crack. Or one that stops the bread coming out today with a shape that drifts slightly from yesterday's.
The bridge to 2026
That's where what I've spent years deploying comes in.
A neural network trained on your own line learns the correct pattern and catches the deviation even though nobody has ever described it. That's the whole difference.
Let me give you a case. At an aerospace plant we applied it to oil leaks on the test benches. They'd spent years chasing that with classical machine vision and with dyes, without getting anything robust — because an incipient drop refuses to be defined with rules, however hard you try. Two months into training, the system was sharper than the human eye.
I see it as the poka-yoke Shingo couldn't build. The same principle as always — don't let the error advance — applied at last to what can't be modelled in advance.