It is very easy to see the benefits that metrics provide (typically you start at no metrics at all), but incredibly hard to measure the costs. This bias is ubi

It is very easy to see the benefits that metrics provide (typically you start at no metrics at all), but incredibly hard to measure the costs. This bias is ubiquitous: metrics, unit testing, abstractions, startup funding, documentation, security, etc. All of those typically start at zero. All of those are perceived as "good", therefore more is always better. You rarely find people with the balls to say "we need less documentation" or "we invested too much effort in security". Kudos to author for recognizing the same issue with metrics. And he's right: the solution is to recognize our bias and for things where cost is hard to measure just assign some cost to it. Though instead of million dollar value suggested by author you should just treat cost as exponential. Having 3 metrics is better than 0. But having 50 metrics is way worse than 3. In fact, having 50 metrics is likely way worse than 0.