The Data Is Right. That's the Problem.

Billy Beane won because he noticed the scouts were wrong.

They said, "He looks like a ballplayer."

The numbers said, "He gets on base."

The numbers won, as they should have. Evidence beat mythology, and baseball got smarter.

And then something less discussed happened.

Other teams learned. They hired analysts. They built models. They started valuing the same things Oakland valued.

And the advantage disappeared.

Not because Billy Beane was wrong. Because he was right.

Correct ideas get copied. And once everybody believes your contrarian insight, it is no longer contrarian.

The data was right. That was the problem.

Who else can see this?

Five companies look at the same market.

They see the same customer behavior. They buy the same analyst reports. They run the same telemetry. And increasingly they point the same frontier model at all of it and ask it roughly the same questions.

They all become extremely intelligent. And they all reach roughly the same conclusion. Same segment. Same opportunity. Same feature. Same roadmap.

Welcome to the red ocean. Everyone arrived scientifically.

This is not a failure of analysis. It is what happens when analysis works and is no longer scarce. Rigor stops being an advantage the moment everyone has it.

I want to make this argument the hard way. So grant the data everything.

Assume it is clean. Assume the model is right. Assume the analysis is airtight.

It still cannot answer the question you are actually asking.

The data answers: is this a good opportunity?

The real question is: is this still a good opportunity after everyone reading the same evidence acts on it?

That is not a measurement. It is a game. And the other players' moves are not in your dataset, because they have not happened yet.

There are four positions.

You can be wrong and crowded.

Wrong and alone.

Right and crowded.

Or right before the crowd.

The first two lose. The third can make money. The fourth creates fortunes.

Because being correct is not the same thing as having an edge. An edge lives in the distance between what you believe and what everyone else believes. Correctness closes that distance. Consensus closes it completely.

Once everyone sees the same opportunity, returns start moving toward the cost of competing for it.

So the question is not "is the analysis right?"

It is "what is still valuable after everyone else realizes it is right?"

Which means the strength of your business case is not the reassurance you think it is. A slide so convincing that nobody in the room objects is a slide your competitors have already built. If it were genuinely non-obvious, someone would be arguing with you.

Ask what agents need, and much of an industry arrives at some version of the same answer.

Context. Structured organizational knowledge. Relationships. Memory. A graph.

And they are right. That is precisely the problem.

If a strategic insight is now visible to an entire industry, the insight itself is no longer the strategy. The strategy is the non-obvious choice you make inside that consensus.

Who else can see this?

Here is where essays like this usually cheat, so let me not.

The track record of conviction over evidence is grim.

Quibi raised 1.75 billion dollars on the belief that people wanted premium short-form video. Segway was going to reorganize cities. Amazon's Fire Phone ended in a charge the company recorded at roughly 170 million dollars, primarily inventory valuation and supplier commitments, disclosed in its own third-quarter filing in 2014.

Bezos had conviction. Katzenberg had conviction. Jobs had conviction.

History does not award points for conviction.

We remember the winners because they won. We never see the thousands who trusted their gut into insolvency, because they left no case studies.

So "trust your instincts" is not a strategy. It is a eulogy waiting for a subject.

The useful question is narrower. When is a trained instinct actually carrying information?

Daniel Kahneman spent much of his career showing why intuition fails.

Gary Klein spent much of his showing when experts somehow just know.

Eventually they worked together to find out why they disagreed. They found they disagreed less than people assumed.

Good intuition needs two things. A world with patterns. And enough real feedback to have learned them.

That is it.

So experience does not automatically create intuition. Experience with feedback does.

Four product launches where nobody ever tells the leader they were wrong is not expertise.

That is tenure.

And one more finding from that work, which is the uncomfortable part: how confident you feel is not evidence that your intuition is any good.

Not predicting the future. That claim has no defense and never has.

Here is the version I think holds.

Every market lives inside constraints.

Something is too expensive. Too slow. Technically impossible. Socially unacceptable. Regulated. Undistributed.

All your historical data describes how people behaved while those constraints were true.

And then one day something changes.

Compute becomes cheap. A regulation disappears. A technical threshold is crossed. Distribution opens.

And now you have a peculiar situation.

The data is still accurate.

It is simply describing a world that no longer exists.

That is where trained intuition earns its place. Not because it can predict the future. Because an experienced person may recognize which old assumption has just stopped being true before the dashboard notices.

That is the actual thesis. Not that Jobs had a better gut. That he noticed when yesterday's constraints no longer deserved to govern tomorrow's design.

The discipline this demands is unglamorous: name the constraint. Date it. Make it arguable. "Inference cost fell ninety percent in eighteen months" is a constraint change. "AI changes everything" is a mood.

And notice what this does to the game. Everyone sharing the same data also shares the same assumption that the current regime holds. Discontinuity is the one thing a consensus cannot price.

Two proposals arrive at the same review.

The first: two billion dollar business, customers requesting the feature, projected 2.7 percent conversion lift. Numbers everywhere.

The second: we think the interaction model itself is wrong.

"How big is the market?" It does not exist yet. "How many customers asked for this?" Almost none. "What is the forecast?" We do not know.

One of these looks responsible. The other looks like a resignation letter.

So we fund the first. Repeatedly. And it compounds. What sells gets investment. What gets investment improves. What improves sells more. The dashboard confirms we were right each time.

We call this being data-driven. What it often is, is letting the past vote on the future. The past has an enormous voting bloc and it turns out in every election.

The Fire Phone did not fail because Amazon made a bet.

It failed because the cost of finding out the bet was wrong was enormous.

That is the separable variable. Not whether you have conviction. What conviction is allowed to purchase before it has earned anything.

A hypothesis should earn the right to consume capital. It should not receive capital because a powerful person believes it.

So the sequence inverts.

Instinct proposes. A specific claim about what changes when a named constraint lifts.

Experiment manufactures the evidence that does not exist yet. Not a survey. Something real, in front of real people.

Data attacks the hypothesis. Mercilessly. Did behavior change? Were we wrong?

Judgment decides. Kill, iterate, or scale.

Four questions before you spend anything.

• Who else can see this? If the answer is everyone, the analysis is a queue ticket, not an edge.
• What constraint am I claiming has lifted? Name it. Date it.
• Have I earned signals here? Enough cycles, in this domain, with feedback you actually received.
• How fast will I know I am wrong, and what does it cost? If the answer is three years and everything, you do not have a hypothesis. You have a mortgage.

So I am not asking anyone to trust their gut.

I am asking for something harder.

Earn an intuition. Name the assumption you believe has stopped being true. Make the smallest real bet capable of proving you wrong.

And then invite the data in. Not to tell you what to imagine. To tell you whether reality agreed.

Because data is extraordinary at answering questions about worlds that already exist.

But every important new thing begins with a different question. What if this world does not have to stay this way?

There was no dataset proving the iPhone. No spreadsheet could contain the first iPhone customer. They did not exist yet. Someone had to build the product before the behavior could produce the data.

That is the paradox. We have built organizations that demand evidence for the future, when sometimes the only way to obtain that evidence is to build a piece of the future first.

So use the data. Respect it. Let it embarrass you. Let it kill ideas you love. Let it make the known world dramatically better.

The data is right.

But when everyone has the same data, everyone has the same map. And maps have one peculiar property. They are most detailed where people have already been.

The empty part is frightening. No benchmarks. No TAM slide. No cohort. No analyst report assuring you that land exists.

Only judgment, a hypothesis, and the willingness to find out.

So before you celebrate the beautiful analysis, ask one more question.

Who else can see this?

Because if the answer is everybody, you may be completely right.

And you may have absolutely no edge.

Use data to navigate the known. But do not confuse the map with the territory still waiting to be found. Occasionally the most important decision a leader makes is to look at an extraordinarily accurate map and decide where not to follow it.

Amazon.com, Inc., Form 10-Q for the quarter ended September 30, 2014. Charges estimated at 170 million dollars primarily related to Fire phone inventory valuation and supplier commitment costs; 83 million dollars of remaining Fire phone inventory and non-cancellable supplier commitments.

Kahneman, D. and Klein, G. (2009). "Conditions for Intuitive Expertise: A Failure to Disagree." American Psychologist, 64(6), 515-526.