I've not heard of pairing metrics before, thanks for mentioning this. It seems like a simple (and obvious in hindsight) way to mitigate the problems that come with focusing on single metrics.

Some brief Googling revealed the following from Marc Andreeson:

Andy Grove had the answer: For every metric, there should another “paired” metric that addresses adverse consequences of the first metric. [0]

Next question for me is, how does one robustly select and pair metrics?

There's a little more here: http://mattyford.com/blog/2014/6/11/paired-metrics

[0] http://pmarcatweetsasblogposts.tumblr.com/post/73631082205/m...