By 1984, according to a 1990 account, graffiti covered every car in the New York City subway, about 6,000 of them. That May, two trains entered a Clean Car Program announced by David Gunn, the Transit Authority’s new president. One ran new cars on the No. 4 line, the other 20-year-old repainted cars on the No. 7.
At the end of each line, a cleaning crew waited. By the same account, a program car that came in tagged had to be cleaned within two hours or it came out of service. Police rode along, and the yards where the clean trains slept got better lighting and cleaners around the clock.
Over five years, the authority later said, trains came out of service in rush hour 10 times. Then, in yards where program cars and older cars were stored together, the people who broke in began painting the cars that were already covered and leaving the clean ones alone.
Proof nobody can wave away
Win a few visible battles early, or many people give up before the change pays off. Failing to plan for those wins is the sixth of eight mistakes John Kotter listed in his 1995 Harvard Business Review article, drawn from watching more than 100 companies attempt transformations. “Most people won’t go on the long march,” he wrote, “unless they see compelling evidence in 12 to 24 months that the journey is producing expected results.”
The win has to be unambiguous, he argued, not “a judgment call that can be discounted by those opposing change.” And it has to be planned, not hoped for. At one manufacturer, the team leading the change picked a new product about six months in because it could be launched fast, by a small team devoted to the new vision, outside the usual departments. Launched about 20 months in, it “boosted the credibility of the renewal process.”
The subway program fit. Two trains were few enough to defend completely, and a clean car was easy to see. It grew only as fast as it could hold, and by the 1990 account beat its yearly clean-car target every year.
Like the rest of Kotter’s eight steps, this comes from his own observation, not controlled studies. A 2012 review by Steven Appelbaum and colleagues found research support for most steps, but no formal study of the whole model. Its popularity, they concluded, appears to come more from its usable format than from scientific consensus.
Why small moves people
The psychologist Karl Weick made a related case in 1984. Problems framed at a massive scale, he wrote, overload people’s reasoning and stir up enough stress to block fresh action. The alternative was the small win, “a concrete, complete, implemented outcome of moderate importance.” One may look trivial, but a series “reveals a pattern that may attract allies, deter opponents, and lower resistance to subsequent proposals.” His paper argues from examples and earlier research; it doesn’t test the idea.
Teresa Amabile and Steven Kramer studied how progress feels. In research they described in HBR in 2011, 238 people on 26 project teams at seven companies filled in a daily diary survey, nearly 12,000 entries in all. Progress was present on 76% of people’s best-mood days. Small steps counted: 28% of incidents with a minor impact on a project had a major impact on how people felt about it. The authors say their analyses show correlation, not causation, and that the progress principle needs work people find meaningful.
The ceremony
Kotter’s seventh mistake mirrors the sixth: declaring victory too soon. “While celebrating a win is fine,” he wrote, “declaring the war won can be catastrophic.” In all but two of a dozen reengineering efforts he watched, victory was declared when the first major project ended after two to three years, and within two more years the useful changes had slowly disappeared. Resisters, he noted, are quick to point to a win as proof the troops can go home.
On May 12, 1989, New York’s last graffiti-covered car was taken out of service and cleaned, and the Transit Authority celebrated. In 1990, hits on trains doubled, and the authority reconvened its graffiti task force for the first time in 18 months. Nobody showed the celebration caused the rise: officials couldn’t explain it, and some writers and a riders’ group blamed budget cuts. Either way, the win had to be defended again.
One shift first
Imagine a nurse manager who wants her ward on a new handover checklist. Her staff have watched initiatives arrive by email and fade. So she starts with the night-to-morning handover, the one everyone complains about, and agrees with that shift beforehand what better looks like. If it works, the day team sees it without being told. The harder moment comes next: the urge to announce success and move on while the habit is weeks old.
A pilot before the rollout
The same logic fits AI. The UK government’s Incubator for AI built Consult, a tool that sorts free-text answers to public consultations into themes. Its first live test was one Scottish Government consultation, on regulating non-surgical cosmetic procedures, with six experts checking the themes on every response. In the team’s 2025 evaluation, reviewers changed nothing for 60% of responses, the differences barely moved the overall ranking of themes, and a median check took 23 seconds.
It has roughly the shape Kotter described: a narrow job, a small team, a result that’s hard to dismiss. The limits are as plain. The team that built the tool ran the evaluation. Consult struggled to spot missing themes, its theme sign-off process “would not scale well,” and reviewers wanted more confidence before checking only a share of responses. The team’s site now says Consult has supported 38 consultations across 10 departments, with a civil-service-wide rollout planned for 2027. A good pilot is the two clean trains, not the clean fleet.