You Are Not Data Driven. You Are Data Justified.

Most people cannot name a single instance. Some go back three or four years to find one. A few find one and then admit the reversal happened because a customer complained, and the data got pulled after the fact to explain the change to the board.

That silence is the whole article, and it costs more than any tool on your stack.

The claim and the receipts

Every company says it is data driven. It is in the values deck, the job postings, the board narrative, and the pitch to the last investor. The spend backs the claim. Warehouses, dashboards, analysts, attribution platforms, a Chief Data Officer, and now a Chief AI Officer sitting next to them.

The people running those functions report that it works. In the 2026 AI and Data Leadership Executive Benchmark Survey, the fifteenth annual edition, 99.1 percent of the roughly 110 large companies surveyed called data and AI investment a top organizational priority. And 97.3 percent said they deliver measurable business value from it.

Hold that number next to another one. MIT’s Project NANDA published research in 2025 finding that roughly 95 percent of enterprise generative AI pilots produced no measurable effect on profit and loss, built from 150 leader interviews, 350 employee surveys, and 300 public deployments. Worth saying, since it will come up: that report is preliminary and has not been peer reviewed, and its outcome measures lean on self reporting. Treat it as one strong signal rather than settled fact.

Even discounted, the two numbers cannot both describe the same world. Ninety seven percent delivering value against ninety five percent producing none.

The gap is not a measurement error. It is a question of who got asked. In the benchmark survey, 90 percent of respondents held the title Chief Data Officer, Chief Data and Analytics Officer, or Chief AI Officer. The survey asked the data function whether the data function is working. That is the same structural flaw I described in the case study piece, wearing a lab coat. The party making the claim is the party being evaluated.

The admission buried in the same survey

There is a finding in that report that nobody quotes, and it gives the game away.

Asked what stands between them and getting value out of data and AI, 93.2 percent of those executives pointed at culture and change management. Only 6.8 percent blamed technology. That is the highest reading on human obstacles in the history of the survey, and the number has sat above 90 percent in four of the last six years.

Translate it. The tools work. The pipelines run. The dashboards load. What does not move is the decision.

The industry reads that finding as a change management problem and responds with more training, more enablement, more adoption metrics. That reading is comfortable and wrong. Nobody needs training to look at a chart. When the same obstacle shows up for fifteen years across a hundred of the largest companies on earth, it is not a skills gap. It is the design of the system. I made a version of this argument in the piece on how good strategy gets killed, where the idea survives review and dies in execution. This is the same machine running one step earlier, on the evidence rather than the plan.

Decision Laundering

Here is the trap, and it deserves a name. Call it Decision Laundering.

Money laundering takes funds with an origin you cannot show anyone and runs them through a legitimate business until they come out the other side looking clean. Decision Laundering does the same thing to judgment. Somebody makes a call from instinct, politics, or the preference of the highest paid person in the room. Then the call goes through an analysis, and it emerges as a finding.

Nothing about the decision changed. Only its paperwork did.

You can spot the shape of it once you know to look. The analysis request arrives with the conclusion already attached, phrased as pull the numbers that show the campaign worked. The date range moves until the line points the right way. Metrics that support the position get promoted to the top of the deck while the ones that do not become context. And nobody asks the one question that would break the spell, which is what result would have made us abandon this.

The output is not knowledge. It is documentation. Your analysts are not informing decisions, they are notarizing them, and the more sophisticated your stack gets the more convincing the notarization looks.

What it costs you

Count it, because this is real money.

The analytics stack, the warehouse bill, the seats on the visualization platform, and the salaries of people producing decks whose conclusions were written before the query ran. Every hour a smart analyst spends building support for a settled decision is an hour they did not spend finding something nobody expected.

Then count the second cost, which is worse and never shows on a budget. Decision Laundering makes you slower at being wrong. A company that reverses course on evidence corrects in weeks. A company that manufactures evidence to support what it already chose corrects when the damage becomes impossible to ignore, which tends to be two quarters and a lot of money later. The dashboard did not just fail to help. It extended the runway of the mistake by making it look examined.

And there is a third cost that hits your best people first. Analysts know when they are being used as a print shop. The good ones ask once whether the answer is open, get a look, and stop asking. What you keep is the ones who never asked.

Run the test. Ten seconds.

Ask one question, out loud, in your next leadership meeting.

Name the last decision our data reversed.

Then be quiet and let the room work. The silence is the finding, and everyone in the room hears it at the same time, which is the part that makes this test different from reading an article about it.

Three answers are possible.

Nobody can name one. Your data function is a print shop and everybody has been polite about it.

Somebody names one and it turns out to be a decision nobody had settled yet. That is data informing a choice, which is good and normal, and it is not a reversal. Reversal means the organization had momentum in one direction and evidence stopped it.

Somebody names a real reversal, with a date, and the room remembers it. Now ask the follow up. How long did the argument take, and who paid a price for having been wrong? If the answer is nobody paid anything, you have a culture where changing your mind is cheap. Protect it, because it is rare, and it is worth more than the entire stack that produced the chart.

The fix is not more data

You do not solve this by buying better tooling. The 93 percent already told you the tooling is not the problem.

Write the kill condition before the analysis. Before anyone opens a query, the person requesting it states in writing what result would make them abandon the plan. If they cannot answer, the analysis is theater and everyone can save the week. This single habit does more than any dashboard redesign, because it forces the decision to be open before the evidence arrives.

Separate the request from the requester. When the person who owns the outcome also scopes the analysis, the scope will bend. It does not require dishonesty, only ordinary human preference, applied to date ranges and segment definitions.

Then keep a reversal log. One page. Every decision the evidence changed, with the date. If the page is empty at the end of the year, you are not data driven, and you have a document that proves it, which is more than most companies have.

Being data driven was never about how much data you collect. Any company can buy that. It is about whether anything you find is allowed to win an argument it was not invited to. Most organizations have never tested this, because the question never gets asked out loud, and the spending continues either way.

So ask it this week. Ten seconds, one question, and the answer will tell you more than the next quarter of reporting.