Data-driven decisions still need judgement

Nobody ever lost an argument by saying the data told them to

It has become almost impossible to challenge a decision once someone describes it as data-driven. The phrase ends the conversation. It signals rigour, objectivity, a clear head unclouded by ego or politics. To question a data-driven decision is to look like the person who prefers gut feel and anecdote, which no senior leader wants to be. So the phrase does quiet work in meetings. It moves a decision beyond scrutiny by attaching it to something that sounds like fact.

That is precisely why it is worth examining. Data is not the problem. The unexamined virtue is.

What the dashboard cannot do

The case for evidence in decision-making is not in dispute here. Leaders who ignore data in favour of instinct tend to make worse decisions, and the research on overconfidence weighs heavily against that habit. The work of Daniel Kahneman, set out in full in his 2011 book Thinking, Fast and Slow and extended in his 2021 book Noise, is among the clearest accounts of how unaided human judgement drifts, anchors, and varies. None of what follows is an argument for going back to instinct alone.

The argument is narrower and, we think, more useful. Data informs a decision. It does not make one. Somewhere between the dashboard and the call, a person has to decide what the data means, what it leaves out, and what to do about a situation the numbers only partly describe. That act is judgement, and it does not become less necessary as the data gets better. It becomes more necessary, because better data hides its own gaps more convincingly.

Three failures show up repeatedly when judgement is allowed to atrophy under the cover of being data-driven.

Measuring what is easy, not what matters. Most things that are genuinely important in an organisation are hard to measure: trust, the quality of a team's thinking, whether a leader is developing or coasting, whether a client relationship is deepening or quietly cooling. Most things that are easy to measure are proxies for those, at best. The risk is not that the easy measures are wrong. It is that they are present, and the things that matter are absent, so the decision quietly reorganises itself around what is on the screen. What gets measured gets managed, and what cannot be measured gets forgotten.

False precision. A number carries an authority its underlying reality often does not earn. A model that outputs a single figure to two decimal places feels more solid than a range, even when the range is the honest answer. Engagement scores, risk ratings, weighted pipeline values, capability indices: all of them compress a messy reality into a clean figure, and the cleanness is persuasive in a way that has nothing to do with whether the figure is sound. A leader who has stopped asking how a number was built has stopped exercising judgement, however data-driven the decision looks.

Abdicating the call to the dashboard. This is the quiet one. The dashboard becomes a place to hide. If the numbers made the decision, no individual has to own the judgement, and no individual has to be wrong. The appeal is real, because owning a hard call is uncomfortable. But a leader who can always point to the data when a decision goes badly is a leader who has stopped leading. The accountability that judgement carries is not a flaw in the process. It is the point of having a leader at all.

Judgement is the capability, not the technique

The conventional response to poor decisions is to improve the technique: cleaner data, better models, tighter dashboards, more disciplined process. Some of that genuinely helps. But it treats the decision as a calculation that better inputs will fix, when the harder part is rarely the calculation. It is the human being doing the deciding.

This is the same observation that sits underneath our companion piece on decision-making under pressure: the bottleneck in a hard decision is usually not the framework, it is the operator. That article looks at what happens to judgement when stress narrows a leader's thinking in the moment. This one looks at the slower, quieter version of the same problem: what happens to judgement when a leader outsources it to a system over months and years, and the muscle simply wastes.

Three things judgement does that data cannot

Judgement is not the opposite of evidence. It is what you do with evidence. Three functions belong to the person and cannot be delegated to the dashboard.

It decides what the data is not telling you. Every dataset has an edge. Judgement is the capacity to sense where the edge is: which question the numbers cannot answer, which population the sample missed, which behaviour stopped being recorded the moment people knew it was being watched. A leader with developed judgement reads the silence in the data as carefully as the signal.

It holds the number against the situation. Data describes the past in aggregate. A specific decision lives in a particular present, with this team, this client, this constraint, this history. Judgement is what reconciles the general pattern with the particular case, and decides when the case is the exception the pattern does not cover.

It carries the consequence. A model does not lie awake over a redundancy decision. A person does, and should. The weight of a decision, the felt sense of what it costs the people on the other end of it, is information, and it is information no dashboard holds. Stripping it out in the name of objectivity does not make the decision more rigorous. It makes it less complete.

This is why we treat judgement as a leadership capability to be built, not a technique to be installed. Our 4D method is built on the conviction that what holds under load is who a leader is, not the tools they were handed. A leader develops the inner steadiness to sit with an ambiguous number, to resist the false comfort of false precision, and to own a call the data could only inform. That development is the work. The dashboard is just one of its inputs.

What it looks like in practice

The leaders described below are representative composites drawn from patterns we observe in practice, not identifiable individuals.

A retail operations director ran a network of stores against a well-built performance dashboard. The metrics were sound and the discipline was real: weekly reviews, clear targets, fast escalation of any store that dropped below threshold. For two years it worked. Then a region that scored consistently well began losing experienced staff, and the dashboard registered nothing until the numbers it did track started to fall, by which point the cause was months old. The signal had been there the whole time, in things the dashboard was not built to hold: a regional manager whose style had hardened, exit conversations that were never aggregated, a quiet shift in who was choosing to leave. The director had not made a bad decision. He had stopped making decisions, and let the dashboard make them, and the dashboard could only see what it had been told to count. The change that followed was not a better dashboard. It was a director who treated the numbers as the start of a question rather than the end of one, and who went looking for what they did not show.

A divisional head in a financial services firm faced an investment call between two business lines, supported by a model that produced a confident recommendation: one line scored materially higher on weighted projected return. The number was precise and the analysis was competent. What the head noticed, on examination, was that the favoured line scored higher largely because its risks were the kind the model knew how to price, while the other line's risks were real but harder to quantify and had therefore been weighted lightly, almost to nothing. The absence of a number was being read as the absence of a risk. The model was not wrong. It was answering the question it could answer, and presenting the answer as if it were the whole question. The judgement here was not to override the data. It was to see what the precision was concealing, to ask how the figure had been built, to hold the clean number against the messier situation, and to make a call that the model could inform but never make. Both lines were funded, with the second one staged carefully against the risks the model had quietly discounted.

Neither leader lacked data. Both had more of it, and better of it, than most. The constraint in each case was the same: the capability to use evidence as an input to judgement rather than a replacement for it. That capability is exactly what tends to thin out in organisations that have made being data-driven a virtue in itself.

Where your own judgement is thinning

The honest place to start is not with your dashboards. It is with your relationship to them. A few questions worth sitting with, for yourself and for the leaders you are developing.

Reader's self-check

• When was the last time you overrode the data, and could you say clearly why? If you cannot remember overriding it, that is the finding.
• For your three most important decisions this year, can you name what the data could not tell you, and what you did about that gap?
• When a decision goes wrong, does your team reach for the analysis to explain it, or for an owner? One of those is a culture of judgement. The other is a culture of cover.
• Which of the things that genuinely matter in your part of the business are not on any dashboard, and when did you last give them deliberate attention?

If those questions are uncomfortable, that is the signal. The discomfort is judgement noticing it has gone quiet. You can see how we assess and develop this kind of capability in our assessments and across our wider work with leadership teams.

The call is still yours

Being data-driven is not a virtue or a vice. It is a description of one input. The virtue, the part that endures and the part worth building, is the judgement that decides what the data means and then carries the consequence of acting on it. No dashboard relieves a leader of that, and the better the dashboard gets, the more quietly the responsibility tries to slip away.

If you want to look at where judgement is being outsourced in your own organisation, and what it would take to rebuild it, we are glad to have that conversation. You can reach us at /contact.