How to be wrong: measure the mean when you really should measure the median.
Consider a website that reports a mean (average) of 2.1 pages per visitor.
Then realize that the median is
9. (actually, my math nerd friends inform me that my example is impossible, because of the peskiness of zero and negative pages. I'll leave the number here as a testament to my carelessness, and will inform you that my point remains correct... that the distribution is way more important than the average. I hope Godel will forgive me).
That's because there's a large number of people visiting 1 page and a large number visiting 10 or 20.
Once you see that, you will completely change your understanding of what's happening and what you need to do to change it. (In this case, the goal wouldn't be incremental improvement from 2.1 to 2.4. It would be to figure out how to activate the one-page people and turn them into 20-page viewers... a 2000% increase from each impacted person).
You'll find the same behavior among McDonald's customers. The typical (mean) American eats a meal at McDonald's once every two weeks. But I never go and some people eat there twice a day. That's a lot more useful to know.
[Bill did a graph.]