Skip to content
Anil Thapa
Data in the business

Data-driven is a discipline, and it lapses

Almost every company calls itself data-driven and almost none defines it. A definition you can test in a meeting, a five-rung ladder most organizations misplace themselves on, and why the tooling gets bought in the wrong order.

6 min read

“Data-driven” is on the values page of almost every company I have worked with or talked to, and I have rarely seen anyone define it. In practice it means “we have dashboards,” or “we hired a data team,” or “the CEO mentions numbers in the all-hands.” None of those is the thing. They are evidence that data exists, not that it is driving anything.

A definition you can test

An organization is data-driven when it uses data consistently, at every level, to change decisions.

Three conditions. Each one rules out something commonly mistaken for the real thing.

Consistently rules out the quarterly set piece. Data that appears when the board meets and vanishes in between is a presentation habit, not a decision habit.

At every level rules out the executive dashboard as proof of culture. If leadership sees numbers and the people running the funnel, the queue or the campaign do not, decisions rest on data at exactly one altitude, and it is the altitude with the least contact with the customer.

To change decisions is the hard one. It rules out data that gets presented, nodded at, and followed by the decision everyone had already made. A number that confirms the plan counts too, if it could have changed it.

If the room goes quiet, the culture is not there yet, however good the platform is.

The ladder

I describe data culture as five rungs, each named for what happens when a number and an opinion disagree, because that is the moment a culture shows itself. The value is in describing each rung well enough that people recognize their own. Different functions in one company often sit on different rungs, and any of them can slip back under pressure.

  1. 1Quoted when convenientA number appears when it supports the position already held. When one disagrees, nobody asks for it.
  2. 2Funded, not consultedThere is a team, a tool and airtime. When a number and an opinion disagree, the opinion carries and the number is explained away.
  3. 3Depends on the roomSome teams reach for the number first. What happens when it disagrees with an opinion depends on which individuals are present.
  4. 4Argued on the evidenceMost decisions start from the number. A disagreement becomes a question about the definition or the data, and leadership pays for definitions, training and time.
  5. 5Limits knownEvidence informs strategy and routine work. The team can say when the data is insufficient, and goes to find out rather than guessing.
Five rungs of data culture. In my experience, most organizations sit on the second or third and describe themselves on the fourth.

The pattern I have seen repeatedly is that organizations place themselves one rung, sometimes two, above where their meetings say they are. The self-assessment is done by looking at the tooling. The real assessment is done by watching the next disagreement.

Four questions place an organization more reliably than any survey:

  1. When a metric moves, does someone ask what changed before someone proposes what to do?
  2. Do people bring data to a meeting unprompted, or only when asked to justify a position they already hold?
  3. Does the same metric name mean the same thing in two departments’ decks?
  4. When the data contradicts a senior person’s instinct, what happens next, and does it happen the same way each time?

None of these has anything to do with the warehouse.

Built in one order, bought in the other

A data capability gets built in a fixed order: culture, then strategy, then technology. It gets purchased in exactly the reverse.

Built
  1. Culture
  2. Strategy
  3. Technology
Bought
  1. Technology
  2. Strategy
  3. Culture
The order a capability is built in, and the order it is paid for. The gap between the two rows is where most data investment goes to die.

Technology is bought first because it is purchasable. It has a vendor, a price, a launch date and a screenshot for the board. Culture has none of those. It is slow, it is mostly conversations, and there is no moment at which it can be announced as done. So the natural move is to buy the tool and expect the culture to follow, and what follows instead is a tool nobody uses. A small technical win can still earn support for the harder work. It just cannot stand in for deciding which behavior should change.

Self-service analytics is the canonical case. The seat was provisioned and the discipline was assumed. I have written about why in What BI means, revisited and in Four habits of a self-service analyst.

The AI wave is repeating this at higher speed. A chat assistant that answers questions in plain language is the most purchasable data capability ever built, and it is landing in organizations that have not settled what their metrics mean. The interface gets solved and the culture gap becomes visible faster, which is at least an improvement on dashboards, which hid it for years.

Four moves that shift a rung

From the leadership seat, the moves that have actually shifted an organization upward are unglamorous, and none of them is a procurement.

1. Ask for the number before the opinion. When leaders open with “what does the data say” rather than closing with it, the meeting reorganizes around evidence. It costs nothing and is the fastest culture change available, because people prepare for the question they expect to be asked.

2. Make one definition official per metric. Not documented, official. Two departments with two conversion rates is not a data problem. It is a decision problem, because they will make different decisions and both will be right by their own number. If the two rates really measure different things, give them different names. Governance is the boring word for ending this.

3. Make the decision the unit of work. A request for a dashboard is a request for a chart. A request framed as “we need to decide whether to keep this channel” is a request for an answer, and it has a finish line. A data team that insists on the second framing does less work and changes more decisions. I have made the same argument about dashboards, where it applies with more force.

4. Put context and data in the same conversation. The business side knows what changed, when, and why it might matter. The data side knows what the numbers can and cannot support. Neither can do the analysis alone, and most bad conclusions come from one side working without the other. The business brings the half nobody can automate.

What it costs

It is slow. It cannot be put on a roadmap with a delivery date, which makes it hard to fund and harder to show progress on. A leader who commits to it will spend a year on conversations that produce nothing a board can see, and then the meetings will be different, and nobody will be able to point to the quarter in which it happened.

It asks leaders to be wrong in public. Consistency means acting on the number when instinct disagrees, and sometimes the instinct will have been right. A culture where data changes decisions is a culture where decisions can be visibly wrong for a documented reason, and not everyone wants that on the record.

The data team gives something up too. Moving from the third rung to the fourth means the central team stops being the only voice that can speak about the numbers. That is the point, and it still stings.

A mirror, not a badge

You are not data-driven because you have data. You are not data-driven because you bought the tool, hired the team or put the word on the values page. You are data-driven on the days you use a number to change a decision, at whatever level the decision sits, and you are not on the days you do not.

It is a discipline, and like every discipline it lapses the moment you stop practicing it. The ladder is useful as a mirror. It is useless as a badge.