Many capable people know a particular frustration at work. They can see that something is wrong: “We are about to lose this account,” “that supplier may be saying all the right things, but they won’t be able to give us what we need,” “that team will not react to this like you think they will,” “I feel like she can step up and get this done.” They may not always be able to give an immediate, complete account of what they see, but they know it in the way a sailor knows the sea: through an intuition accumulated over many repetitions, over many years. However, their judgment may often be treated as provisional until it can be translated into “data” or another simplistic, flattened representation. The expert may be expected to explain themselves in the language of the representation, and the person who knows the representation is often permitted to speak as though they know the situation. What can be measured, displayed, audited, and transferred has suddenly come to be seen as more real than what is understood through deep immersion. 

Complexity in work grows with the number of elements and their relationships that must be understood. Five relevant elements in a decision create ten potential pairwise relationships, ten create forty-five, and twenty create one hundred and ninety. All of this is before we count higher order relationships. A situation with twice as many relevant factors can impose much more than twice the cognitive work. Going beyond the sheer arithmetic of pairs, relationships have timing, direction, and sequence. A decision by one function alters the conditions under which another function responds, and that response changes the decision-maker’s next options. An intervention intended to solve one problem may create a different problem several months later. The situation is an interacting system whose parts continually alter one another. At lower levels of work, a capable person may be able to hold the important variables and their relationships directly. At higher levels, work becomes the organization of relationships across people, functions, time horizons, and consequences. Data can report the elements and some of their movements, but it cannot, by itself, determine which relationships are structurally central in this particular situation, or what sequence of action the whole now requires. That is a fundamental limitation of data, not just a temporary issue with measurement. 

Despite that, data is often spoken of as though it places you within the situation by itself. It does not. A dataset is a representation: a selection from the situation, shaped before analysis even begins. Decisions have already been made about what counts, what can be counted, where the boundary of the problem is, when measurement will occur, and which categories will stand in for a more dynamic and complex reality. The analysis begins after these decisions, not before them. Most data is a reasonable attempt to bring into view something that would otherwise remain unseen. But it is a compression, not the thing itself. Its value depends on how accurately it records what it includes, and on the discipline with which we understand the limits of what it omits. A data-based decision is therefore a modest and worthwhile claim: the decision-maker has made a good-faith effort to examine the fragments of the situation that can be measured, rather than decide without looking at them at all. The data does not resolve the situation, but it prevents some forms of blindness.

The notion of “data-driven,” however, has gained a life of its own. This notion makes a much stronger, and often unexamined, claim by granting those fragments of reality, i.e., data, a higher rank than other forms of knowledge and making data an admission ticket for credibility. The phrase has essentially acquired a moral force and become a modern virtue. A decision described as data-driven sounds rigorous and responsible even if nobody has asked what was measured, what was omitted, or whether the metric has the depth needed to drive a decision. 

Why should this partial way of knowing command more authority than the person who must act within the whole situation? Data is useful, but usefulness cannot explain the degree of deference it now receives. We do not simply consult it anymore, we ask it to make our judgments legitimate. The phrase “data-driven” carries the appeal of an impersonal guarantee: the decision appears to come from the world itself, rather than from a person who has had to interpret the world. Perhaps the world becomes more bearable when it feels solved. A number, a dashboard, or a model does not merely describe something, it gives your perspective an edge. It turns what is dynamic and uncertain into something that can be placed before a group, compared, passed on, and acted upon. In that movement, the unrepresented begins to feel like an intentional, thoughtful omission - something that ought to have been captured if it mattered. Judgment, on the other hand, offers no such relief because it is always personal to some extent. The person who judges must say, in effect, “this is what I see, this is why I think it matters, and I take responsibility for acting based on this.” That burden may be one reason data is prized. It may also be why organizations prefer knowledge that can be taught and transferred to knowledge that has to be formed slowly in someone. Whether the motive is anxiety, fatigue, an appetite for control, or something else, data worship promises a way to escape the discomfort of complexity without actually having mastered it.

Once this reversed hierarchy is in place, deeply immersed judgment is often forced to disguise itself. The person who has learned to recognize a pattern through years of contact must search for a number, a process, or a model that will allow the judgment to be received. Some translation is useful, of course, because it can expose a judgment to challenge, discipline it, and force the expert to say more clearly what they are seeing. But it becomes an issue when the translation is treated as the source of the judgment rather than as evidence that bears upon it. The effect is, naturally, a reversal of authority - the person who knows the terrain must defer to the person who knows the map. Someone fluent in representation can appear more credible than someone with more actual contact with the situation, simply because their knowledge arrives in a form the organization knows how to recognize. As a result of this, the expert’s sense of pride can take a hit, and the organization mistakes what it can transfer quickly for what it actually needs. The earned knowledge becomes increasingly difficult to notice, value, and form through apprenticeship and trust - two things a culture of legibility tends to regard with suspicion.

“Data-driven” should lose some of its moral glamour. It tells us something about the source of a decision, but nothing by itself about the depth of the situation, the quality of the interpretation, or the capacity of the person who must act. The more serious “virtue” is to know what the data can show, what it cannot show, and what actions or responsibilities an analysis of it is leading to. A person who has spent years learning to recognize a domain should not need to pretend that their perception emerged from a metric in order to be taken seriously. Nor should they be exempt from the obligation to test their perception against evidence. The standard is neither certainty without data nor submission to data. It is practiced, accountable judgment: the willingness to see as much as possible, admit what cannot yet be seen, and still take responsibility for action.

The world does not become less complex just because it has been made legible. The key to good work is to meet complexity where it is, without pretending that it has already been solved for.