Lean tools

Andon

In French: Andon

The andon is the device that makes an anomaly visible the instant it occurs, and calls for help. It is not a warning light: it is an alert chain whose value lies in the response time.

A signal, a response, a time

Andon is Japanese for a paper lantern. In a workshop, the word covers the whole device: the means of triggering — a button, a cord, a call — the means of signalling — a light column, a screen, a bell — and above all the response it triggers.

That third element is the only one that produces an effect. A light column nobody comes to look at is a decorative object; the same equipment combined with a commitment to intervene within a few minutes changes the way the line works.

The device is therefore judged on a time, not on its installation: how long between the trigger and the arrival of someone who can act. It is the only measure that distinguishes a living andon from an abandoned one, and it is easy to record.

The alert levels

A useful andon distinguishes several situations, otherwise everything is equal and nothing is urgent any more. Three levels are almost always enough: a request for assistance without stopping, an anomaly that will stop production if nobody intervenes, and an actual stoppage.

Each level has a different recipient and a different time. Confusing the three produces either over-solicitation of management, or the habit of not responding — two outcomes that end in the same place, a device nobody triggers any more.

The colour and position of the signal must be readable from the place one intervenes from, not from the station. It is a common layout mistake: the signal is perfectly visible to the person who has just triggered it, and invisible from the supervisors’ office.

What the device produces once running

It turns endured problems into treated problems. An anomaly reported nowhere is absorbed by the person who meets it: they improvise, they catch up, and the information never leaves the station. Triggering gets that information out and makes it available for improvement.

It also provides a datum few workshops have: the frequency and nature of calls, station by station. That list is a ready-made order of priority, built by those who suffer the problems rather than by those who assume them.

It finally shifts the role of management, and that is the hardest change. Answering a call then takes precedence over meetings and dashboards, which assumes that the hierarchy has explicitly decided so and stands by it over time.

The common mistakes

Installing the hardware before having defined who responds, within what time and instead of what. It is the reverse of the right order, and it produces a visible investment whose use dies out within a few weeks.

Tracking the number of triggers as an indicator to bring down. It is the best way to obtain an immediate and misleading fall: people stop calling, the problems remain. What is steered is the response time and the resolution rate.

Reserving the trigger for management or the most senior. The device is worth something only if it is within immediate reach of the person who observes, with no permission to ask.

On the shop floor — Automotive assembly

The board existed and worked, but calls were rare. Observation showed that the station had to wait a quarter of an hour on average before a setter became free, and that the lost production was counted against the station. The operators had concluded, without anyone telling them, that calling cost them more than coping.

The other tools illustrated by this trade: automotive.

Frequently asked questions

Can an andon be transposed outside industry?

Yes, and the principle rarely loses its value there: in a department, the equivalent is an immediate alert channel combined with a response commitment, often carried by a shared tool. What counts is not the medium but the same three elements — a means of signalling within reach, a designated recipient, and a time that is kept.

How do you measure whether the andon works?

By the time between the trigger and the arrival of help, recorded over a period, and by the proportion of calls that lead to a lasting action rather than a mere restart. A device whose response time lengthens month after month is dying, even if the number of calls remains stable.

What should be done if there are too many calls?

Treat them as a diagnosis rather than as a nuisance: a station that calls constantly points to a real problem the organisation was silently absorbing until then. The answer is to solve the most frequent causes, not to restrict access to the trigger — restricting simply brings back the previous silence.

Should the line be stopped at every call?

No, and that is precisely the point of the levels: most calls request assistance while production continues. Stopping is reserved for situations where carrying on would produce bad parts or put someone in danger. Confusing call and stop makes the device so expensive that nobody dares use it any more.