Why improvements come undone, and what actually holds a gain

Decay is the normal behaviour of an unattended change

A new method competes daily with the old one, which is easier, familiar and better suited to the pressures of a bad shift. Each small deviation is individually reasonable: skipping the staging step because the previous job overran, borrowing the dedicated fixture because another machine needed it, missing a check because somebody was absent. None of these is misconduct and none is visible on its own. Sustaining is therefore not a matter of insisting harder. It is a matter of arranging things so the deviation is noticed quickly and somebody whose job it is asks about it.

A change without a named owner has already begun to unwind

Improvement events end with actions assigned to people who then return to their normal jobs. The change itself, meaning the new standard, the tooling, the board and the routine, needs a permanent owner in the line organisation rather than in the improvement function, and that ownership should appear in what the person is asked about by their manager. Where the improvement team retains ownership, the area treats the change as belonging to somebody else, complies while watched, and reverts afterwards. Handover at the end of an event is a specific step, and it is routinely skipped.

The routine that checks the routine

Layered confirmation means the team leader verifies a small number of critical process points frequently, the area manager verifies a subset less often, and a senior manager samples occasionally, each looking at whether the standard is being followed rather than at output. Its value lies less in catching non-compliance than in surfacing standards that cannot be followed as written, which is what most failures turn out to be. It works only where a finding leads to fixing the standard or the conditions rather than to a conversation about the operator, and where senior managers actually perform their layer.

Turnover quietly removes the change

New starters learn the method from whoever is next to them, and the method they learn is the one being practised rather than the one documented. A cell with steady staffing can hold a change for a long time on memory alone; the same cell after several leavers, a shift pattern change or a period of agency cover will have drifted substantially. The countermeasures are unexciting. The standard has to be the induction material, somebody competent must confirm the new person can perform it, and periods of high turnover deserve more confirmation rather than less.

People do what they are measured on, eventually

If a supervisor is judged on machine utilisation, the smaller batches introduced by a flow project will grow back. If the plant reports overhead absorption, work in progress will return. If nobody is asked about changeover times, they will lengthen. The most reliable way to hold a change is to alter the conflicting measure and the routine question at the same time as the change itself. This is a leadership decision and the single most common omission, because changing reported measures is harder and much less satisfying than improving the process.

Frequently asked questions

Why do our improvements disappear once outside support leaves?

Because that support was supplying the attention the management system does not. Consultants and internal improvement teams run the routines, chase the actions and ask the questions; when they go, nothing in the ordinary working week performs those functions. The test to apply before any engagement ends is which of the ongoing routines are being run by line managers as part of their normal job. If the honest answer is none, the gains will unwind regardless of how good the work was.

How do we confirm compliance without it becoming policing?

Ask about the standard rather than about the person, act on what you find, and let the finding change the standard whenever the standard is wrong. If every visit produces a correction aimed at the operator, people will present compliance rather than reality and the exercise becomes worthless. If a reasonable proportion of findings result in the process or the documented method being fixed, the same visit is experienced as support. The difference lies entirely in what happens after the observation.

What should we do when the standard turns out to be unworkable?

Change it quickly and visibly, through whoever owns it, and tell the area it changed because they said so. Standards known to be wrong that remain in force teach everybody that documentation is decorative, and that lesson generalises to every other standard in the plant. A fast, low-ceremony route for revising a work standard is worth more than any amount of auditing, and its absence is the usual reason obsolete instructions stay in the folder for years.