# The music business doesn't have a data problem. It has a context problem.

By Zack Massey · November 8, 2026 · https://www.zackmassey.com/writing/context-problem

> Not more data. Everybody has dashboards, and the arguments in the room are just as long. People rarely disagree about the number. They disagree about what it's evidence of, and that context lives in people's heads and leaves when they change jobs. AI made the analysis cheap. Knowing when a confident answer is wrong is still the scarce part.

Everybody has dashboards now. That is the part that actually changed.

Fifteen years ago the argument in most rooms was that we needed more information. We got it. There is now more information about a song in its first week than there used to be about an artist in a year. And the arguments in the room are exactly as long as they used to be. They just have charts in them.

I have sat in a lot of those rooms. Radio, label, management. The thing I keep noticing is that the number is almost never what people are disagreeing about. Two people look at the same curve. One says the song is working. The other says the song is being pushed. They are both looking at the same chart, and neither of them is wrong about the chart.

What they disagree about is what the chart is evidence of.

That is a context problem, and no amount of additional data touches it.

## What context actually means

Context is everything that was true around the number.

A record adds forty stations this week. Is that good? It depends on which forty. Whether they were already in on the last one. Whether somebody called in a favor that costs something in six months. Whether the adds are in markets the artist can actually get to. Whether the same forty adds happened for somebody else in the spring and went nowhere.

None of that is in the number. All of it is in somebody's head.

And that is the real shape of the problem. The context exists. It is just distributed across people, and it decays. The person who knew why a station made a call in March has changed companies by August. The number stays in the system with none of the reason attached to it, and a year later somebody builds a model on it.

## Why more data made it worse

Here is the part I did not expect.

When there were three numbers, everyone knew they were working off almost nothing, so they argued about the thing itself. Someone in the room had been doing it for twenty years and their read got weight, because there was nothing else to give weight to.

Now there are three hundred numbers, and the numbers feel like the argument. So the experienced person's read gets treated as an anecdote. It gets called a gut call, which in most rooms is now a polite way of saying you do not have a slide.

But that read was not a feeling. It was compressed context. It was fifteen years of watching what happened after this exact pattern, and the fact that it cannot be sourced does not mean it is not evidence. It means we have not figured out how to keep it.

That is the loss nobody is measuring. We replaced a small amount of information with a lot of it, and along the way we quietly downgraded the only thing that told us what any of it meant.

## And now anybody can ask

The thing that changed recently is that the analysis got cheap.

Somebody can drop a spreadsheet into a chat window and have a competent read in about forty seconds. I do it. It is genuinely useful and I am not going to pretend otherwise while building with the same tools.

But it moved the bottleneck rather than removing it. The model does the arithmetic correctly and then tells you a story about the arithmetic, and the story is where the trouble is. It does not know that the bump in week three was one station in Dallas going into heavy rotation. It does not know the artist was on a stadium run through the south that month. It does not know that the record it is quietly comparing this one to had a completely different setup behind it. It gives you a confident answer built on the numbers you handed it and nothing else, because the nothing else was never in the file.

That is not a complaint about the tools. It is a description of what they have to work with.

What it does mean is that the scarce thing moved again. It used to be information. Then it was analysis. Now it is being able to look at a plausible, well written, internally consistent answer and know that it is wrong.

## What I think this actually is

I do not think this is a tooling problem, exactly, and I am suspicious of anyone who tells you their product solves it. Most of what gets sold as a solution here is another dashboard, which is more of the thing that caused it.

What I think is closer to true: the industry got very good at capturing what happened and stayed bad at capturing why anybody thought it happened. Those are different jobs. The first one is a database. The second one is harder, because the answer lives in a person, changes over time, and is sometimes wrong.

I do not have a clean answer. I have spent a while now building against the edges of it, and the honest position is that the interesting work is in the second job and almost all of the money has gone into the first.

The tell, if you want one: next time somebody presents a number in a meeting, count how long it takes before someone asks what it means. If the answer is instant and everyone agrees, you did not need the number. If the answer takes twenty minutes, the twenty minutes was the valuable part, and nobody wrote any of it down.