Feedback for Asymmetry, Not Just Stability

Definition

Classical control-system thinking, inherited from manufacturing quality control, treats feedback as a mechanism for reducing deviation from a static target — an "inverted U-curve" payoff where you profit by staying near the peak. Reinertsen argues product development's payoffs are asymmetric and its goals are dynamic, so feedback should be designed differently: fast feedback lowers expected loss by truncating unproductive paths before they compound, and raises expected gain by letting you redirect resources quickly toward an emergent opportunity. The same feedback loop pushes on both tails of the distribution at once.

In the Book

Chapter 8 opens by warning readers with a manufacturing background that they will need "a significant mental shift": manufacturing feedback loops chase static goals with symmetric payoff-functions, where variation always hurts. Product development's payoff-functions are the asymmetric ones described in the book's variability chapter, so fast feedback is valuable for a different reason — it doesn't just reduce variance, it reshapes which parts of the outcome distribution you actually experience. Principle FF3, "The Principle of Leading Indicators," gives a concrete mechanism: one manager told Reinertsen he tracked task start times rather than completion times, because a late or wrong-resourced start reliably predicted a late finish, and by the time a completion date slipped, intervention was already too expensive. Principle FF1 and FF2 add the economic filter for choosing what to monitor at all: among development expense, queue size, and cycle time, queue size wins as a control variable because it combines high economic influence with being a leading (not lagging) indicator — you can act on a growing queue before it turns into a blown schedule.

Why It Matters

This distinguishes two different jobs feedback can do, and conflating them leads to mis-designed control systems: reducing variance around a known target (right for stable, repetitive processes) versus reshaping an uncertain outcome's distribution by cutting losses short and extending gains (right for any exploratory, asymmetric-payoff activity). It also supplies a general method for choosing what to measure in any domain — rank candidate metrics by economic influence, then by how efficiently and how early each one can be acted on — instead of defaulting to whatever is easiest to count.