North Star metrics lie

Published on Sep 10, 2020

There's no North Star metric. North Star metric is a fancy and abstract term. It either doesn't exist or isn't universal enough. It's impossible to reflect any aspect of a product with one number. The world is more complex than that. North Star metrics are rather a result, a composition of everything. They are handy in talks about business performance and operating at a high level. But they cannot say how the product works. They are also useful in some particular cases. This limitedness makes them not universal, and thus, not a North Star.

A simple example is LTV - customer lifetime value, or how much money on average you get from one user. Comparing it with cost per user you can calculate how profitable the product is and track its growth. And that's it. Moreover, most of the North Star metrics are averages. The problem with averages is that they do not tell the whole story. What is true for the group isn't true for the sub-group or individual. In LTV case, If the average goes up it means you're making more money on average. But at the same time, the total number of users you have may go down and total revenue it will be less than before.

Tracking the North Star metric is like looking through a telescope. You clearly see your goal but if you do it too long you may miss anything that happens at your feet. You need a set of metrics that requires a profound understanding of all business aspects. Everything from UA to product, from the technical side to design and UX is important. One minor thing can get broken and everything will collapse. You'll never see this with only one metric. Each star in your constellation of metric should tell the exact story. You should understand how your actions and external factors influence it. You should know what happens if it goes up or down and how it's distributed across the userbase.

Too many metrics is also bad. Don't overdo with them. When you put too things into your dashboards, you can get lost in the woods. You'll be led by correlations instead of causations and by random noise. You should always understand WHYs behind metrics' changes and all WHAT IFs. Being reasonable and considering potential biases will help you to be fully armed.