It's not that ARPU is meaningless, it's just averages by themselves can be easily misleading the same way percentages by themselves can be misleading.

> It turns out that for most businesses, revenue per customer follows a power law distribution.

It would have been better to provide a source for that statement, the way the author provides data for Movies, Cities, and Web Traffic.

Precisely. This piece seems like an over reaction to the discovery that averages are a relatively blunt way of looking at data. They're not terribly useful in a vacuum. Looking at a distribution (histogram) of your data is a great way to understand the distribution of your data, but even then, you have to be careful about how you bin your data.

This is good advice for data analysis in general. When looking at data, I like to start with a histogram. Web server benchmarking is a great example. Developers appear to have a strong desire to boil things down to a single number.

"What's my endpoint response time?"

The answer is often an average, but the data is rarely well represented by that number. Looking at a histogram reveals that.