A leader can make a process worse while trying very hard to improve it. That is the danger of tampering. When every disappointing result triggers a new rule, a new explanation, or a new adjustment, management may be reacting to routine variation as if something unusual happened.
The result is more noise, more frustration, and less learning. The better question is not, What changed yesterday? The better question is, Do we know what kind of variation we are seeing?
The temptation to chase the last number
W. Edwards Deming used a simple funnel experiment to make this problem visible. Imagine dropping a marble through a funnel toward a target. Even if the funnel stays in the same position, the marble will not land in exactly the same place each time.
The natural impulse is to move the funnel after each miss, trying to compensate for the last result. That feels sensible. It feels active. It feels like control. But if the process is already stable, repeated adjustment can spread the results farther from the target.
The act that feels like control becomes a source of instability. Management can fall into the same pattern whenever it treats the latest result as a command.
A dashboard turns red. A customer complains. A weekly number dips. Someone asks for an explanation, and the organization rushes to change the work. Sometimes that response is necessary. A real special cause deserves attention. But when the process is stable, the better work is to improve the system, not chase each point.
That is the problem ClearStep, a mid-sized B2B software company, faced when its support leaders tried to improve response time by changing the process after every bad day.
The support dashboard that would not settle down
ClearStep sold project management software to manufacturers. Its support team handled setup questions, bug reports, billing issues, and urgent support calls. The team was capable, but its work arrived unevenly.
Rina, ClearStep’s head of customer support, watched one number more than any other: median first response time. When it rose, customers complained. When it fell, the executive team relaxed.
Monday morning, the dashboard looked bad. Response time had jumped from twenty-three minutes to thirty-seven. Rina opened the team meeting with a decision already forming.
“We need a new rule. For the rest of the week, no one works on follow-up tickets until the new queue is under control.”
Marcus, the operations analyst who helped the support team study workflow data, hesitated. He had been plotting daily response time for the past six months.
“I know thirty-seven minutes looks bad,” Marcus said. “But it is still inside the range we have seen before.”
“Customers do not care about ranges,” Rina said. “They care that we were slow.”
“Agreed,” Marcus said. “But if we change the rule every time the number moves, we may be adding variation ourselves.”
That was not what Rina wanted to hear. She was trying to be responsive, not careless. The team had already changed the escalation rule twice that month. One week, senior agents took every urgent ticket first. The next week, new tickets came first.
By Thursday, response time improved, but reopenings were up. Customers got quick replies that did not resolve the issue. The team was moving faster and learning less.
Deming named the trap plainly: “Mistake 1. To react to an outcome as if it came from a special cause, when actually it came from common causes of variation.”
Mistake 1. To react to an outcome as if it came from a special cause, when actually it came from common causes of variation.
— W. Edwards Deming
Rina asked Marcus to show the chart again. The bad Monday was unpleasant, but it was not outside the usual pattern. The system had been predictable for months. Response time bounced within a wide band because of uneven ticket routing, inconsistent urgency definitions, and too few agents trained on integration issues.
“So doing nothing is the answer?” Rina asked.
“No,” Marcus said. “Studying the system before we change the rules is the answer.”
“Then what do we change?”
“Not the queue every morning. We change the conditions that keep creating these wide swings.”
That distinction changed the conversation. ClearStep still investigated real signals: outages, product releases, unusual customer spikes. But it stopped rewriting queue rules after ordinary variation. Rina’s team clarified urgency definitions, cross-trained agents on integration questions, and reviewed blocked tickets each day to remove causes of delay.
The solution was not inaction. It was action aimed at the system.
Why we keep treating noise like a signal
We drift into tampering because the pressure to respond is real. A leader sees a bad number and feels responsible for it. A customer is waiting. A team is anxious. An executive wants an explanation. In that moment, studying variation can sound like delay.
But the demand for an explanation can create its own distortion. If every up and down requires a story, people will supply stories. Some will be true. Some will be guesses. Some will be shaped by what seems safest to say. The organization may become better at explaining variation than reducing it.
Deming’s warning is uncomfortable because it challenges a common picture of leadership. We often equate visible reaction with accountability. We expect the manager to change something, tighten something, or call someone into the room. But if the latest point came from common causes, the visible reaction may make the system harder to understand.
Deming put the problem sharply: “They were tampering with a stable system, making things worse.”
This is not only an internal efficiency problem. A company that keeps changing priorities teaches employees to protect themselves from the latest swing. It teaches customers to expect inconsistency. Over time, reliability becomes harder to deliver, and trust becomes harder to earn.
The management habit that looks decisive in the moment can quietly weaken the system that customers experience.
What leaders can do instead
Before choosing a response, leaders need a way to separate movement that calls for investigation from movement that calls for system improvement. Deming was not arguing for passivity. He was arguing for action that fits the evidence.
Observe the system before reacting. Do not treat the latest point as a command. Look at performance over time and ask whether the process is showing a real signal or behaving as it has behaved before.
Separate urgency from interpretation. A customer problem may need immediate care, but that does not mean the process itself needs a new rule. Serve the customer, then study what the result means.
Ask better management questions. Instead of asking, “Who caused yesterday’s miss?” ask, “What conditions keep producing this range of results?” The second question moves attention from blame to capability.
Improve the sources of variation. Work on definitions, training, handoffs, equipment, priorities, and decision rules. These are less dramatic than a new order from management, but they are usually closer to the real causes.
Build reliability as a management advantage. When a company reduces unnecessary variation, people can plan, customers can trust the service, and leaders can learn faster. That consistency is hard to copy because it comes from the way the system is managed.
The key is to avoid confusing energy with improvement. Activity can make management feel involved while leaving the process worse than before.
Better leadership starts by seeing variation
Tampering is tempting because action feels responsible. But leadership is not measured by how quickly a manager changes something after a bad number. It is measured by whether the action fits the evidence.
Some results call for investigation. Some call for patience and system improvement. The skill is knowing the difference. When leaders stop chasing noise, they create the conditions for steadier work, clearer learning, and better performance tomorrow.
That is the practical promise of understanding variation: better action, taken for better reasons.
