The Art and Science of Data-Driven Decision Making
Kurosawa's *Rashomon* tells the same story four times from four perspectives, and the film ends without telling you which version is true. That's what most data looks like in a contested room. The same numbers support multiple readings. The reading that wins is the one the most persuasive person makes most plausibly. Pretending otherwise, pretending the data speaks for itself, pretending the chart resolves the question, is the central lie of "data-driven" culture. The honest move is to admit the data is evidence and the decision is judgment, and to take responsibility for both.
"Data-driven" is a phrase that's been hollowed out by repetition. Every company says it. Every leader endorses it. Almost nobody means the same thing by it, and the gap between the slogan and the actual practice is where most of the bad decisions get made. So I want to write about what data-driven decision-making actually looks like when it's working, and what it looks like when it isn't, based on the rooms I've been in.
The first thing to say is that data, by itself, decides nothing. It's evidence. The decision is what someone does with the evidence, and the someone matters as much as the evidence. The same dataset, presented to two different teams, produces two different decisions. Not because one team is right and the other wrong, but because data acquires meaning through the questions you bring to it, and different questions produce different conclusions from the same numbers. Anyone who has watched two analysts argue about a chart knows this is true. The chart is the same. The argument is real.
The second thing to say is that the most data-driven cultures I've seen up close are not the ones with the most dashboards. They're the ones where the people in the room have spent enough time with the data that they don't need to look at the dashboard during the meeting. The dashboard is for the people who don't yet know the data. The decision happens between people who already know it. When a meeting starts with "let me pull up the chart," the decision is going to be slower and worse than when the meeting starts with "based on what we saw last quarter, here's what I think we should do." Internalized data beats displayed data, every time.
The third thing is that "data-driven" gets weaponized in ways that have nothing to do with making good decisions. Sometimes the phrase means "I have evidence and you don't, so I win." Sometimes it means "the analyst's interpretation is final because they have the spreadsheet." Sometimes it means "any decision that can't be defended with a number is irrational and should be reversed." All three of these are bad ways to run a company, and all three are common. The cultures that use data well treat it as one input among several. The cultures that don't treat it as a trump card.
I've been in rooms where data was used well. They share a few features. The decision-makers know the data already, so the meeting is short. The analyst is treated as a thinking person with a perspective, not as a calculator with a presentation. Disagreements are about interpretation, not about whether the data is "right." Someone in the room is allowed to say "I think the data is misleading us here, and here's why," without being treated as anti-evidence. The decision, when it happens, references the data without being controlled by it.
I've also been in rooms where data was used badly. Those rooms share different features. The decision was already made before the meeting, and the data is being marshaled to defend it. Or the data is being used to delay a decision indefinitely because nobody wants to commit until the dashboard is "complete," which it never will be. Or the most senior person reframes the data to support their prior view and the room goes along because nobody wants the fight. The data, in all these cases, is theater. The decision was always going to happen the way it happened. The numbers were the staging.
A few patterns I've come to believe:
The right number of metrics on a dashboard is small. Three to five. Maybe seven. If you have twenty, you have a database, not a dashboard, and the people looking at it will not make better decisions than they would have made without it. Pick the metrics that move with the decisions you actually make, and accept that the others are noise.
The dashboard is not the analysis. The dashboard is the surveillance system that tells you when something needs analysis. Most teams confuse these. They build elaborate dashboards and call it analytics work, when really they've just instrumented the surface and never gone underneath. The interesting work is in the underneath. The dashboard tells you something changed. The analysis tells you why, and what to do about it.
Beware the precision trap. A model that says revenue will be $4,372,118 next quarter sounds more credible than one that says "between three-eight and four-six." The second one is more honest. The first one is performing certainty it doesn't have. Most business decisions need the range, not the point. Forecasts that hide their uncertainty produce decisions that overcommit to the wrong thing.
Calibrate the people, not just the models. The decision-makers who use data well are the ones who have a track record of being right when they're confident and wrong when they're not, and who know it. The ones who use data badly are confident every time, regardless of outcomes, and never go back to check whether they were right. Calibration is a learnable skill. Most organizations don't measure it. The ones that do produce better decisions over time, not because the data improved, but because the people interpreting it improved.
Expect the data to be late. By the time something shows up clearly in the metrics, the underlying thing has been happening for a while. Lagging indicators lag. Leading indicators are noisy. The teams that wait for the data to be unambiguous are usually the teams that miss the move. The teams that act on partial signal are sometimes wrong but on average earlier, and earlier compounds.
The deeper thing about data-driven decision-making is that the phrase contains a category error. Decisions are not driven by data. They're informed by data, shaped by it, sometimes constrained by it, but the decision itself is always made by a person, applying judgment to evidence that never quite covers the question being asked. The cultures that internalize this make better decisions because they take responsibility for the judgment. The cultures that don't keep pretending the data made the call, which means nobody is responsible when the call turns out to be wrong, which means the same mistake gets made again.
Charlie Parker told a young player once that if you don't live it, it won't come out of your horn. That's the part of data-driven decision-making the dashboards can't capture. The judgment that turns numbers into action comes from the decisions you've made before, the calls you've gotten wrong, the patterns you've watched play out long enough to recognize them when they're starting to play out again. The data is the horn. The lived experience is what comes out of it. Pretending the horn plays itself is how organizations end up surprised by outcomes that anyone with judgment would have seen coming.
