A ceramic and metal joint with a blue face sits on a wooden base between two open viewing frames.

What we can know from an outcome

An outcome can force us to reconsider a decision without explaining, on its own, what happened. Between perception, numbers, and judgment, we need to distinguish what we have learned from what now seems certain.

Alejandro Mayorga12 min read

The outcome of a decision changes the position from which we look at it. Something that was once a possibility has now happened; the alternatives that remained open now belong to a history that never took place. From this new position, some arguments gain a force they did not have when the decision was made, while others begin to seem difficult to defend. There is a legitimate reason for this shift: we know more. The difficulty lies in establishing what we know and how much it changes what we could previously defend. An outcome may reveal a significant omission, confirm a risk we had recognized, or still leave unresolved why things happened as they did.

The distinction matters because an outcome makes it easier to bring an activity to a close than to settle on an explanation. A delivery was completed, a deadline was missed, an intervention coincided with an improvement. The record of what happened may be quite clear. Establishing the relationship between that record and the quality of the judgment that preceded it requires further work. When the two are confused, the conclusion seems to come built into the event: those who succeeded understood the situation; those who failed should have chosen differently. That convenience comes at a cost. It removes the need to reconstruct what could be observed, which alternatives existed, and how much of the outcome actually depended on the decision.

In The Fixation of Belief (opens in a new tab), Charles S. Peirce examines this inclination to move out of doubt and arrive at a belief on which we can act. Among the methods he considers for fixing belief are persistence in a conviction, acceptance of authority, and scientific inquiry. Their differences matter because each establishes a different relationship with reality. His philosophical argument helps articulate a discomfort that remains relevant: feeling that a question is settled and having good reasons to settle it are experiences that may coincide, but there is no guarantee that they will. The relief an explanation brings tells us little about its quality.

What changes after a decision

Jonathan Baron and John Hershey studied a specific part of this problem in their experiments on outcome bias (opens in a new tab), published in 1988. Participants received descriptions of decisions made under uncertainty and were told their outcomes. Although they were instructed that they had the relevant information available to the decision-maker, they rated the reasoning more favorably when the outcome was positive. The study shows how an outcome can influence judgment even when the information on which the decision was based is described. A preregistered replication published in 2023 (opens in a new tab) found the effect again in a hypothetical medical scenario. Its scope remains limited: neither these scenarios nor their participants represent the full complexity of evaluating decisions within an organization.

It would be strange to use this evidence to conclude that outcomes should no longer matter to us. In real work, we rarely know in advance all the relevant information, the quality of the estimates, and the actual reasons behind people’s decisions. Some of what we know comes precisely from acting and observing the consequences. An outcome may reveal a condition that was missing from the analysis. It may also show that a capability attributed to a team was not available in the circumstances in which it was needed. Giving up that information for fear of bias would shield some interpretations from the possibility of correction.

Consider a hypothetical decision to commit to a delivery date. The team has a reasonable estimate and has identified an external dependency that could affect the deadline. The dependency fails, and delivery is delayed. That outcome alone does not show that the estimate was poorly constructed: a recognized risk can materialize. But pointing out that the risk appeared in the analysis is not enough either. We would still need to examine whether it was reasonable to accept it, whether its significance was adequately represented, and whether those who committed to the date had the authority to pass its consequences on to others. Having anticipated a possibility does not automatically justify having accepted it.

The interpretation changes if an investigation into the delay uncovers evidence that this dependency had already been failing repeatedly and that the estimate had omitted available information. There is now a different reason to reassess the judgment. The outcome led to the discovery of a weakness in its foundation. We need to preserve that sequence: what requires a change in the assessment is the knowledge gained by examining what happened, not merely whether the outcome was favorable or unfavorable. Without that distinction, two decisions reached in very different ways may receive the same criticism because both ended badly.

Some outcomes also justify revising an explanation without making the person who relied on it culpable. A forecast may have been defensible and prove inadequate in light of new information. Correcting it for the future does not require pretending that this information was always available. Responsibility for a past decision and willingness to keep using its assumptions are related questions, but they refer to different points in time. An organization can recognize that someone acted reasonably and, at the same time, conclude that repeating the same choice would no longer be reasonable.

That possibility becomes difficult when changing one’s mind is interpreted as admitting that one acted wrongly before. The person who needs to revise a hypothesis then has reasons to defend it beyond what the new evidence supports. At the opposite extreme, every adverse outcome may prompt a new rule, even when there is still no adequate explanation of what happened. Between defending the previous judgment and replacing it in haste lies a less conspicuous task: establishing which part of the interpretation has lost its support. The outcome begins that review; it rarely completes it.

The authority we grant a number

Quantification can help with this work. It allows us to recognize differences that are hard to notice, test expectations, and examine whether an impression persists as observation broadens. Its strength depends, however, on our understanding of the relationship between what we count and what we want to understand. Delivery time describes one dimension of work. A rating of coordination attempts to capture another. Expressing both as numbers does not make them interchangeable or mean that they automatically answer the same question.

A measure brings together prior decisions about what to observe, how to record it, what to compare, and what to leave out. The mathematical operation may be carried out correctly while one of those decisions proves unsuitable for its subsequent use. Averaging ratings from people in different working relationships produces an average; to know what it represents, we still need to examine what each source was able to observe. Someone receiving a delivery and someone involved in preparing it may know different aspects of the coordination. The distance between their judgments may contain information about the work, as well as differences in criteria. Erasing it to obtain a more convenient figure may remove precisely what we needed to investigate.

Hands use an outside caliper to check the thickness of an irregular stone on a workbench.

Recognizing this mediation does not put every interpretation on the same footing. A specific observation, supported by identifiable situations and checked against relevant sources, offers better reasons than a general impression that cannot be examined. Perception can be well grounded while remaining situated. A number can provide a valid comparison without describing the phenomenon in full. The requirement is to preserve the relationship between the evidence and the claim, so that we can discuss where an interpretation gains strength and where it begins to exceed what was observed.

Statistics has developed tools for working with this uncertainty; borrowing its appearance to conceal uncertainty undoes part of its contribution. In its 2016 statement on p-values (opens in a new tab), the American Statistical Association cautions that scientific conclusions and decisions should not depend solely on whether a value crosses a threshold. A p-value does not express the probability that a hypothesis is true, nor does it measure the importance of an effect. These clarifications belong to statistical inference and should not simply be transferred to an organizational score. They do, however, help us recognize an error of interpretation: granting a numerical output the power to settle a broader question than the one it was designed to address.

A statistical conclusion requires design, assumptions, and interpretation. An interpretation of perceptions requires criteria, exposure, and context. Their requirements are not identical, and treating them as though they were would produce another form of false precision. The confidence stated by someone making an assessment is not equivalent to a statistical measure of uncertainty either. It may help us understand that person’s relationship with the evidence, but it does not acquire the properties of a confidence interval because the two expressions share a word. Conceptual precision begins before calculation and continues after it.

This discipline is especially important when a figure enters a decision that has consequences for others. The need to approve, allocate, or intervene may require a definite answer even when the available knowledge remains limited. It does not follow that we should disguise those limits. Someone with authority can decide with acknowledged uncertainty and take responsibility for the consequences. If the figure is presented as an answer sufficient in itself, that responsibility becomes harder to locate: it seems as though the decision was contained in the data and someone merely carried it out.

An interpretation open to correction

In work on organizational capabilities, we need to preserve the distinctions between outcomes and perceptions. An operational improvement deserves attention; we still need to understand what contributed to it and under what conditions. A favorable interpretation of how a capability manifests itself may also be relevant, even though it does not, on its own, demonstrate a better business outcome. When one source silently takes the place of the other, the assessment is credited with conclusions it did not produce, or aggregate performance is used as a reason to stop examining how the work is functioning.

PRYSMAP is subject to that requirement. Structuring, contrasting, and contextualizing perceptions makes sense if it allows us to build interpretations that can be examined. That proposition loses its foundation when the score is ultimately presented as a truth about the person or as sufficient proof of a stable capability. Traceability matters because it allows us to revisit the interpretation: what was observed, from which relationship, against which expectation, and with what limits. The value of retracing that path includes the possibility of discovering that the initial interpretation needs to change.

A person examines the joints on the back of a carved wooden panel resting on sawhorses.

The relationship with a client takes on greater seriousness when that possibility is part of what is offered. Delivering a conclusion accompanied by a general warning about its limitations is not enough. We need to be able to explain which claim is supported, which decision it is relevant to, and where further evidence would be needed. If an organization asks whether its team is ready to take on new work, a capability assessment provides part of the answer. The specific demands, the conditions of execution, and the consequences of being wrong remain part of the decision. No format should make those questions disappear to make the diagnosis seem more conclusive.

Nor would it be enough to turn caution into language that never commits to anything. An interpretation open to revision needs to assert something precisely enough for experience to put it to the test. If any subsequent outcome is compatible with what we said, perhaps we did not say very much. Openness to correction requires us to recognize which observation would weaken our explanation, which finding would strengthen it, and which aspects remain beyond its scope. That requirement applies both to the organization making a decision and to those who provide it with tools to support that decision.

An outcome, then, deserves something more demanding than obedience or distrust: it deserves to be questioned in relation to the explanation it is supposed to support. Sometimes it will reveal an error; at other times, it will confirm that we were acting under a risk we could not eliminate. Sometimes we will still not know enough to distinguish between the two. The uncertainty that remains does not invalidate the knowledge gained, but it sets the limits of what we can answer for. An important part of the responsibility of deciding lies there: in standing by what is well founded and letting what happened compel us to understand more fully, even when that understanding takes away the confidence with which we had spoken.

Taking responsibility for a decision includes being willing to lose the certainty with which we made it; that demand applies to us at PRYSMAP too.

References