19.7 Decision Correction

Decision correction is the process by which a decision-making system identifies that a prior decision has produced or is producing undesirable outcomes and modifies its subsequent choices to reduce the gap between actual outcomes and intended goals. It is the adaptive core of cybernetic decision systems: the negative feedback loop through which errors in decision making are detected, diagnosed, and addressed. Without decision correction, a system that makes an initial mistake has no mechanism for improvement — the error propagates unchanged through subsequent decisions. With effective decision correction, a single poor decision becomes a learning event that improves the quality of subsequent decisions and prevents the accumulation of systematic error.

The Mechanism of Decision Correction

Decision correction proceeds through several distinct stages:

Error detection is the recognition that outcomes are deviating from intended targets. This requires active monitoring: the decision maker or the system must observe outcomes, compare them to expected or desired values, and identify when the comparison reveals a significant deviation. Error detection is dependent on the quality of the feedback system — how quickly, accurately, and completely outcomes are measured and reported back to decision makers. Poor feedback systems generate delayed, distorted, or absent error signals, allowing errors to persist undetected.

Error diagnosis is the identification of what caused the detected error. An outcome can deviate from target for many reasons: the decision was based on incorrect information about the state of the world; the causal model used to predict the effects of the decision was wrong; the decision was correctly conceived but poorly implemented; external conditions changed after the decision in ways that altered its effects; or some combination of these. Without accurate diagnosis, correction efforts are misdirected — adjusting the wrong element of the decision process while the actual source of error persists.

Response selection is the choice among possible corrections. When diagnosis identifies the source of the error, the decision maker must determine what change will address it. If the information base was wrong, the correction is to improve information quality. If the causal model was wrong, the correction is to revise the model. If implementation was defective, the correction addresses implementation. If the error was an anomaly rather than a systematic pattern, no structural correction may be required — the decision rule itself is sound, and the poor outcome reflects an unavoidable random deviation.

Correction implementation is the actual modification of decisions, decision rules, information gathering, or implementation procedures. Correction is not complete until the change is actually made — identified errors that are diagnosed but not addressed leave the system as vulnerable to the same failure as before.

Single-Loop and Double-Loop Correction

Decision correction can occur at two levels, which have substantially different effects on the decision system:

Single-loop correction adjusts decisions within the existing decision framework — changing parameter values, refining estimates, or modifying the specific choices made — without questioning the underlying rules, assumptions, or goals that guided the original decision. If a pricing algorithm generates returns below target, single-loop correction adjusts the algorithm's parameters to produce higher returns without questioning whether profit maximization is the appropriate objective or whether the model structure correctly represents market dynamics. Single-loop correction is appropriate when the decision framework is sound and only specific calibrations are wrong.

Double-loop correction revises the decision framework itself — the goals, assumptions, rules, or models that generated the original decision — rather than merely adjusting parameters within the existing framework. If analysis reveals that the pricing algorithm consistently underperforms because it rests on an incorrect model of competitive dynamics, double-loop correction replaces the model rather than re-tuning parameters within it. Double-loop correction is required when the source of error lies in the framework rather than in its parameterization, but it is more disruptive, more cognitively demanding, and more organizationally difficult than single-loop correction.

The distinction between these two levels of correction is important for diagnosing why organizations fail to learn effectively. Organizations often excel at single-loop correction — they adjust, refine, and optimize within their existing frameworks — while systematically failing at double-loop correction, which would require questioning foundational assumptions that may be politically or psychologically protected.

Barriers to Decision Correction

Several factors make decision correction more difficult than the conceptual description implies:

Sunk cost bias causes decision makers to maintain or double down on failing strategies because of prior investment rather than correcting them. The sunk investment becomes a psychological barrier to acknowledgment that the decision was wrong, leading to escalation of commitment rather than correction.

Attribution ambiguity makes it difficult to determine whether a poor outcome reflects an error in the decision or simply a run of bad luck within a sound decision process. When causal attribution is unclear, decision makers may incorrectly infer that a sound decision produced a bad outcome through chance, and avoid correction that would actually have been warranted.

Organizational defensiveness arises when acknowledging decision errors carries social or political costs. In organizations where mistakes are punished rather than treated as learning events, decision makers have incentives to deny, minimize, or conceal poor outcomes rather than acknowledging them and initiating correction. This defensive dynamic prevents errors from reaching the detection stage of the correction cycle.

Overcorrection occurs when decision makers adjust too aggressively in response to detected errors, overshooting the correct calibration and producing errors in the opposite direction. Overcorrection produces oscillation: the system swings back and forth around the correct setting without converging on it, because each deviation triggers a correction that generates a new deviation in the opposite direction.

Learning from Decision Correction

The most durable benefit of effective decision correction is not the elimination of the specific error that triggered it but the improvement in decision processes that prevents similar errors in the future. When corrections are accompanied by learning — when the reasons for the error are understood and used to improve decision rules, information processes, or causal models — the correction event strengthens the decision system against future failures. When corrections address only the immediate error without learning — when the system adjusts today's decision but does not change the process that generated the error — the same category of error will recur under similar conditions.

Organizations that institutionalize learning from decision corrections — through after-action reviews, systematic post-decision outcome tracking, and explicit processes for incorporating lessons into decision rules — build up a stock of error-proofing over time. Those that treat each correction as an isolated incident, without connecting it to prior patterns or using it to update decision processes, forfeit this cumulative learning benefit.