The Funnel Experiment is Deming's second great teaching demonstration, paired with the-red-bead-experiment as the twin pillars of his seminar pedagogy. Where the Red Bead Experiment demonstrates the futility of holding workers accountable for system-caused variation, the Funnel Experiment demonstrates what happens when management responds to that variation with "corrective" adjustments — a practice Deming called tampering.
The physical apparatus is simple: a marble dropped through a funnel onto a sheet of paper with a target marked on it. The marble comes to rest at some distance from the target; the question is what the operator should do with the funnel's position before the next drop. Deming formalized four response rules, each corresponding to a common management behavior:
Rule 1 — Leave the funnel alone. The funnel stays fixed over the target regardless of where each marble lands. This is the statistically correct response to a stable common-cause system. The marble's scatter around the target is irreducible noise, and no adjustment can reduce it.
Rule 2 — Compensate by the last error. After each drop, the funnel is moved opposite to the last result by the distance the marble deviated. If the marble landed 2 centimeters left, move the funnel 2 centimeters right. This approximates a feedback control system. The result: scatter roughly doubles compared to Rule 1. The well-intentioned correction amplifies variation.
Rule 3 — Move the funnel over the target based on the last error. After each drop, the funnel is repositioned directly over the target, adjusted for where the marble landed relative to target. Unlike Rule 2 (which accumulates positional memory), Rule 3 resets to the target each time. This produces somewhat less instability than Rule 2 but still generates more variation than Rule 1.
Rule 4 — Move the funnel over where the last marble landed. The funnel is repositioned directly above the previous marble's resting point, as if trying to reproduce the last result. This rule produces the most dramatic outcome: the marble's position walks randomly across the paper with no tendency to return to the target, generating a pattern that expands without bound. Deming would watch the marbles drift to the edge of the paper to illustrate this explosive instability.
Each rule is an analog to a real management behavior. Rule 2 describes a manager who issues corrections based on each week's results without understanding baseline variability. Rule 3 describes a manager who resets to targets after each deviation. Rule 4 describes "on-the-job training" in which each trainee learns from the last trainee — a game of telephone that accumulates and amplifies distortion with each generation. Rule 4 is also the logic behind worker training that relies on experienced workers passing on habits without reference to documented process standards: each iteration drifts further from the original.
The Funnel Experiment makes the concept of tampering visceral. Tampering — making adjustments to a stable system in response to common-cause variation — is not mere inefficiency; it actively worsens performance. Management interventions that feel like corrections are revealed as the source of the instability managers are trying to fix. The demonstration connects directly to common-cause-vs-special-cause-variation: the first skill in understanding variation is recognizing whether a system is stable. If it is stable, Rule 1 is always correct. Only when there is evidence of a special cause — something outside the system's normal distribution — should adjustment be made.
Deming treated the Funnel and Red Bead experiments as complementary. The Red Bead Experiment shows that variation exists and that workers cannot overcome system-caused variation by trying harder. The Funnel Experiment shows what happens when management responds to that variation: they make things worse. Together they form the core of the statistical argument in out-of-the-crisis and in the four-day-management-seminars, and they are the experiential foundation on which statistical-process-control-and-variation-theory and the broader system-of-profound-knowledge are built.