One thing I see far too rarely in choosing metrics is ensuring that every metric has one or more countervailing metrics to balance the choices made. In other words, if you are optimizing for a metric, you also need to measure what is being sacrificed for it. The classic trio of cheap/fast/good is a great example of this, as are most engineering-style tradeoffs, like capacity vs range.

You will often see managers create metrics for the things they want to improve, without realizing that making a metric without measuring the countervailing metric is creating an incentive to throw the countervailing metric in the shitter. A good example of this is call centers which measure the time of call resolution without measuring customer satisfaction with the call. This is how you end up with a low quality, high volume call center. You can do the opposite by measuring only satisfaction and not speed. Now, maybe what your business needs is high volume, low quality, but that should be a choice you make. The problem with not measuring the countervailing metric is that nobody (officially) cares about how bad that metric gets. So, if you have a tradeoff situation, you automatically get an extreme even though that may not be the most profitable place on the spectrum. Perhaps your call center would be better for the company if it was volume 9 quality 2 instead of volume 10 quality 1, but if you only optimize for volume you can't make that choice.

This even applies to metrics you really want to go up, things like sales and profit. A manufacturing company that optimizes only for sales while ignoring manufacturing metrics will hire salespeople in preference to manufacturing workers until the inability to deliver product starts affecting sales. That might sound like a nice problem to have, but it can be a real problem, and perhaps sales 9 manufacturing 2 instead of sales 10 manufacturing 1 would result in better sales next year. It's better to have metrics which allow you that choice.