The Tyranny of the Average

Mar 22

A number landed on the board presentation screen. Everyone relaxed. That was the problem.

I'm on the board of Meow Wolf. An 8.5 landed on the board presentation screen and made everyone in the room happy.

Visitor satisfaction, out of 10. It sat there looking excellent and we moved on. I didn't push back. The room was confident., 8.5 felt like something to celebrate, and I kept my mouth shut.

What I should've done was ask what was hiding behind it.

When we broke the data apart, the picture was more complicated. Younger visitors were off the charts. Meow Wolf's immersive, disorienting experience was exactly what they'd come for. But a serious number of visitors had come wanting to follow the exhibit's narrative, to understand the story, to move through it with their families in a way that made sense. And tThey couldn't. They felt lost in a different way than the experience intended, and frustrated in a way that had nothing to do with the art.

It got averaged away.

For a company whose entire proposition is that the experience has to land across generations, that's not a footnote. That's a strategic question. And the average had given the board permission to skip it.

I've made versions of that mistake more times than I'd like to admit. Forty years in the data business and I still have to remind myself: the average isn't the answer. It's the beginning of the question.

The average isn't even the only number doing this. Means, medians, engagement rates, conversion rates, net promoter scores all share the same flaw: they compress complicated reality into a single figure and invite you to stop there. A median tells you where the middle person sits but nothing about how far the edges stretch. An engagement rate folds together your most passionate followers and the people who clicked once by accident and never came back. Single summary numbers are sedatives. They make complexity feel resolved when it isn't.

Digital dashboards make this worse, not better. Platforms generate enormous amounts of behavioral data: clicks, watch time, scroll depth, purchase frequency, all of it averaged into reports that feel precise. Your most valuable customers and your most at-risk ones are buried inside those numbers, behaving completely differently from each other. When you average across all of them, both signals vanish at the same time.

A handful of power users with outsized engagement can pull your average way up, masking the fact that most users are barely engaged at all. A small number of very large transactions can make average order value look strong while most customers are actually spending less than last year. The average won't surface either problem. It'll just sit there looking fine.

Standard deviation exists precisely for this moment: to show you how spread out reality actually is behind that tidy figure. A small standard deviation means most of your data clusters near the average. A large one means the average is a fiction and the real action is somewhere else entirely. It's not a complicated tool. It just requires someone in the room willing to ask for it.

Maimonides, the 12th century philosopher and physician, wrote that wisdom lives between the extremes, but you can't find it until you've honestly understood the full range first. Nobody in that board room was looking at the full range. We were looking at 8.5 and calling it done. The average had answered a question nobody actually asked, and we'd all accepted it.

The follow-up question is the whole job.

Averages earn their place for quick orientation, for trends over time, for comparisons at scale. But they're a starting point, not a destination. The distribution, both ends of it, tells you what's actually there.

Every time a clean number lands on your screen and everyone in the room relaxes, that's your cue. Not to relax, but t. To ask what it's hiding.

The board was happy with 8.5. I should've asked who wasn't.