The Analysis Paradox

That the challenge lies not in collecting data but in analyzing it meaningfully, no one can argue against that. It’s reasonable. But it’s only half the truth, because after collecting and analyzing, someone has to actually do something with it. And that’s the step that’s usually missing.

In the companies I’ve worked with, it mostly went the same way. There was enough data and enough analysis, and every Monday morning we sat in a team meeting looking at a PowerPoint with numbers that were already well known. Customers are leaving, satisfaction is dropping. Everyone in the room sees it and nobody draws any consequences from it. That has nothing to do with the quality of the analysis but with the culture in the company.

It’s often presented as if data analysis were the bottleneck. If only we had better tools, we could finally use the data. AI is supposed to help make things visible that were previously hidden. That sounds good, but it assumes that the company also acts once something becomes visible. And most of the time nothing happens, even though everything is already on the table.

The reason is that action has consequences. Whoever takes the analysis seriously has to change processes and structures, sometimes even replace people. That’s uncomfortable and creates resistance, so instead the next analysis gets commissioned. The analysis becomes a substitute for the decision.

I know a company that uses three different AI tools for customer analytics. The results come in quarterly and are well prepared. In two years, not a single structural change was made based on these analyses. The tools work, but nothing happens anyway.

What almost never comes up in this discussion is the political dimension of data. In companies, data is also an instrument of power. Whoever has the right numbers gets budget, whoever presents the wrong ones loses influence. Analyses aren’t always done to find something out but to support an existing position. AI doesn’t change that, it just makes the tools more precise.

I witnessed a department commission an AI-based analysis and then ignore the result because it didn’t fit the picture. A different time period was chosen, a different model and a different data set, until the numbers confirmed what had already been decided. The tool worked flawlessly, it was just used backwards.

The honest question would not be how we analyze better, but why we don’t do what we already know. This question is rarely answered, because the answer wouldn’t be a project but leadership and the willingness to make unpopular decisions. As long as analysis remains a substitute for decision-making, better tools won’t change anything about that.