kaizen in practice · situational guide
D-KAI-05 | First published 2026-07-18 | Revised 2026-07-18 | Primary sources: 5
Disclosure: Gemba Works Lab is supported by advertising. This page contains no affiliate links; if any are added in the future they will be clearly marked (per the FTC Endorsement Guides).
Somewhere in your plant there is probably a faded banner, a binder of value-stream maps nobody has opened since the consultant left, and a crew that rolls its eyes at the word "lean." This guide is for the owner or plant manager standing in front of that history and wondering whether a second attempt is worth the political capital. The research question we kept in front of us while writing it: what, specifically, dies when a lean program dies — and which of those causes of death can a smaller, quieter restart actually avoid?
KEY POINTS
Before restarting anything, write one page on why the last program stopped. Not the diplomatic version — the mechanical one. The published evidence gives you a checklist of suspects. In the 50-manufacturer study cited above, researchers grouped fifteen documented lean challenges into four categories: culture (43.65% of the challenge weight), technical knowledge (26.45%), management (15.34%), and employees (14.67%); the top individual barriers were lack of management support (16%), lack of key performance indicators (12%), and unstandardized processes (10%)[1]. Note what is missing from that list: wrong consultant, wrong toolset, wrong Japanese vocabulary. Programs rarely die because 5S was taught badly. They die because nobody with authority kept showing up after month three.
The kaizen-event literature reaches the same verdict from the other direction. Fieldwork across 65 events in eight manufacturers found that what happened after each event — management review, institutionalized follow-up routines — was what separated sustained results from decay[2]. The EPA's public training module on kaizen events builds a whole third phase around this, with monthly follow-up meetings written into the method itself[3]. If your autopsy page says "we ran good events and the gains evaporated," you did not have an event problem. You had a phase-three problem.
Match your program's death to one of the seven failure modes catalogued in our pillar guide and name it in writing. The restart plan below assumes you know which one killed you, because each one implies a different first repair.
A failed rollout is not a total loss, and treating it as one wastes real assets. Keep three things. First, any baseline data that was honestly measured — old changeover times, defect counts, and cycle studies are still valid history, and they shorten the measurement month of a restart. Second, the people who were trained. If two leads sat through a five-day lean course, that knowledge did not expire with the program; a documented MEP restart pattern is exactly this — training a small cadre, then aiming them at bounded events. Third, the physical improvements that survived. A shadow board that is still accurate after two years is evidence of what sticks in your plant; study it before building anything new.
Throw away, without ceremony: the branding, the steering committee, the plant-wide metrics dashboard, and the vocabulary that now carries a smirk. If "lean" is a punchline on your floor, stop saying it. The word was never the mechanism. What OSHA's worker-participation guidance identifies as the working ingredient — people engaging with a system that visibly acts on what they report[4] — does not require any particular label to function.
Weeks 1–2: repair the specific breach. Each failure mode has a matching first act. If the last program harvested improvement gains as layoffs, the restart begins with a written, spoken, and kept pledge that improvement never costs employment — and nothing else happens until leadership is willing to sign it. If suggestions went unanswered, the restart begins with a supervisor calendar block, because the 72-hour response rule is a staffing commitment before it is a policy. If events produced gains that drifted back, the restart begins by scheduling the monthly review meetings first, before any improvement activity exists to review[2][3].
Weeks 3–6: one cell, quietly. Pick a single work area — ideally one whose crew was least burned by the old program — and run the baseline month from the pillar guide's 90-day plan, compressed: one metric, measured and posted, plus a paper suggestion form for that cell only. No kickoff meeting, no banner, no all-hands announcement. The absence of launch theater is itself the message that this attempt is different. The structured-coaching pattern NIST highlights in its Toyota Kata discussion — short, repeated, routine interactions rather than sponsorship campaigns[5] — is the model: the restart is a habit being installed, not a program being launched.
Weeks 7–8: first visible wins, publicly credited. Approve the small, cheap, reversible suggestions in batches and post the count of implemented ones. Crews burned by a dead program watch one number: whether things they proposed actually changed. That is the participation mechanism OSHA's guidance describes[4], and it either restarts here or the whole attempt stalls here.
Day 60: a deliberately unglamorous review. One page — baseline, current number, implemented suggestion count, hours spent — and a decision to extend to a second cell or hold. Format and honesty rules are in the pillar's go/no-go section. Whether the second cycle should be another concentrated event or a deeper daily routine depends on cadence trade-offs covered in Kaizen Events vs Daily Teian Systems (D-KAI-08).
Sign 1: the layoff wound is open. If people lost jobs in connection with the last program and leadership has never addressed it plainly, no method survives that memory. The pledge has to come first, from the owner, in person.
Sign 2: no supervisor has two hours. The teian routine costs a supervisor two to three hours a week, indefinitely. If the honest answer is that nobody has them, fix supervision load first — a restart that starves in the same way as the original teaches the crew that nothing will ever change, which is worse than not restarting.
Sign 3: nobody trusts a number. If the old program inflated savings claims, your measurements are politically contaminated. Rebuild trust the slow way: measure one boring, verifiable thing (changeover minutes, not dollars) and let the crew check the stopwatch. Estimates get labeled as estimates — the same discipline we apply to our own cost figures (D-KAI-06).
One more caution: if the deepest problem exposed by the failed rollout is that one senior person carries the whole process in their head, the restart is a knowledge-capture project before it is an improvement project — start with the 30-minute skills matrix (D-SKT-02) and the Skills Transfer section.
Modest and specific: one cell measurably better and holding at its monthly audits, a second cell started with the first cell's operators as coaches, a suggestion board where the implemented count rises most months, and a review meeting that has met every month without exception. No banner. If that reads as a low bar, recall the arithmetic of the alternative — the study above suggests only about one in four North American lean implementations reaches satisfactory results[1], and the ones that do are distinguished by exactly this kind of unphotogenic persistence[2]. The full week-by-week machinery, budgets included, is in The Complete Kaizen Implementation Guide (D-KAI-P01); the vocabulary reset, if your crew needs the seminar version replaced with the shop-floor version, is in What Kaizen Actually Means (D-KAI-01). Note the division of labor between this article and its companion: this page is about recovering after a program has already broken down; our peer-reviewed piece on why kaizen gains disappear after 12 months is about why gains fade even in programs that never visibly "failed" — the quieter, more common version of the same problem.
Last updated: 2026-07-18 | Primary sources referenced: 5 | Figures without a source are labeled as estimates or as our editorial synthesis. Spotted an error? Tell us via the contact page.