Why improvement gains disappear within a year

Improvement gains disappear because they live in habits, and habits drift back. A gain survives only as long as somebody maintains the current best way of doing the work, and that maintenance is the part most teams skip. The workshop gets a celebration; the upkeep gets nobody.

Then the slide starts.

Where does an improvement really live?

Not in the slide deck. An improvement is the gap between the old way and the new way, and the gap exists only while people keep choosing the new way. Once the person who drove the change moves on, the new method starts competing with muscle memory. Muscle memory usually wins. It has home advantage.

Panopto ran the numbers on this with YouGov back in 2018, surveying 1,001 US workers. The finding that sticks: 42 percent of institutional knowledge sits in a single person’s head. Your best process fixes live inside that number. If the better method exists nowhere except in the memory of whoever invented it, you haven’t improved a process. You’ve improved a person. People leave.

Running Tallyfy since 2014, we’ve watched the same arc repeat: a fix ships and works, then fades once its inventor stops policing it. Nobody decides to abandon it. Remembering just stops being anyone’s job.

Put numbers on the decay

One kaizen research roundup is blunt about the scale: 70 percent of improvements decline within 12 months, and undocumented fixes start reverting in about six months on average. Six months. That’s how long a hard-won gain lasts once nobody owns the standard behind it. Meanwhile some team two floors up reinvents a wheel this team already built, because the invention was never written anywhere findable.

There’s a rhythm to how this plays out. Weeks one through eight after a kaizen event look great, because the people who built the change are still in the room and still proud of it. The reversion starts when attention moves to the next event, which is precisely when everyone stops measuring.

This might sound backwards, but the fastest way to lose an improvement is to declare it finished. Finished means unowned. The gain becomes an orphan, with no one checking whether the work still matches the method, and orphaned methods erode at whatever pace turnover and busy weeks set for them.

Figures like these deserve a bit of caution, mind you. Decay rates differ by industry and by how you define a gain in the first place. The direction never differs, though. Ask anyone who has run back-to-back kaizen events what remains of the ones from three years ago. The answer is usually painful, and it says nothing about whether the original fixes were good.

Do teams forget on purpose? No. Forgetting happens because remembering was never assigned.

Standard work is the boring half of kaizen

Back up a step. Taiichi Ohno’s production system at Toyota had two halves, and everybody remembers the exciting one, the improvement events. The half that made gains stick was duller: standard work. Write down the current best known method. Everyone follows it until a better method replaces it. Then the standard itself changes, and the cycle continues from a higher floor. Boring, and load-bearing.

The standard is what makes improvement cumulative instead of decorative. Without one, three things break at once: you can’t say what changed, you can’t train the change, and you can’t tell when it quietly un-changes. Masaaki Imai built much of his 1986 book on kaizen around that same dependency, and the lean world has repeated the lesson ever since.

Which is a polite way of saying most binders die young.

If the standard lives in a binder, it is already losing

A binder can’t tell you it’s stale. Neither can a wiki page. Paper and pixels hold the words of a standard, but they can’t see whether anyone follows those words, and they can’t feel the moment the real work drifts somewhere new. The watching part of maintenance is exactly the part a static document can’t do. We wrote more about that failure in why your SOPs age faster than anyone admits.

The alternative is making the standard the same thing people execute. When each step of a process runs as a tracked task, every run either confirms the standard or exposes it, and drift shows up as data instead of folklore. A living standard can answer things a binder never could, like who followed it last week and where they stalled. It can show which instruction keeps producing the same clarifying question, too. Those answers are the raw material of every improvement worth keeping.

We built Stern Stella on that premise: every version of every step is saved forever, the current standard stays visible, and a proposed change has to carry evidence from real runs before anyone applies it. Kept gains also compound, and the compounding math grows faster than intuition expects.

One misconception we see constantly: teams treat gain decay as a people problem and prescribe more training. Custody is the real problem. The method was never separated from the person who invented it, so it left when their attention did.

Sort of unfair, when you think about it. The team did the hard part, they found the better way, and what failed them was storage.

None of this needs a lean consultant to start. Pick one process that was painful to improve. Find where its current best method is recorded, if anywhere. If the answer is a document last touched in 2023 or a veteran’s head, you’ve found the leak, and you can decide on purpose who owns that standard from now on.

The teams that keep their gains are simply better at remembering. An improvement is an event, retention is a system, and the system decides whether the event mattered.