Not to pike too much, since measuring is a big step from not measuring... but IMHO, this isn't analytics. Graphing and looking at things going up and down is just tracking and reporting. Looking at causes, looking for correlations (when you can't find a cause, see if the correlations give a clue), grouping up data to understand what is "normal" so that real deviations stand out from noise, pattern detection, linking of variables into directional chains (where possible), all that is analytics. (And yes, he does do some manual analysis when one of his metrics is awry... but that doesn't make it analytics, just another step in the right direction.)

I'm as guilty as the next guy to call metrics "analytics", partially because it takes some analysis to pick "the right" metric. But it's gone too far: I fear that we've corrupted analytics (which is really just a corruption of analysis, or maybe a portmanteau?) into being applied to pretty much any dashboard, visualization, or counting system.

We finally have cheap storage, fast counters, great parallel systems... so let's do measure everything, and revel in the data. But let's also admit that simply measuring is not (yet) analytics, just a necessary first step... and that it's just reporting until that additional level of, dare I say it, analysis, is applied to it.

I say this partially to push us all to take that next step, to not accept just counters, even if they are pretty and realtime, but to push for more analytics embedded in our tools, be they easy UI to show interactions between variables through to stats/AI/ML which finds patterns, shows outliers, and reveals opportunities or pitfalls.

Excuse the shameless plug, but if you're interested in metrics and analytics, maybe you'd be interested in joining the AWS CloudWatch team? http://news.ycombinator.com/item?id=2284954

http://xkcd.com/833/ ;-)

"Hi John"