The anomaly detection I actually wanted

Mean time to resolution (MTTR) is the battleground for customer loyalty. But, it's really hard to reduce MTTR for the "unknown unknowns". A new approach is needed to root-cause and sometimes even to detect novel incidents. Anomaly detection has been implemented by various tools and with varying degrees of success, but doesn't rise to the bar of novel incident recognition.

I've spent the last few years developing a new approach that uses machine learning to automatically catch and characterize novel software incidents. I will discuss the approach I've taken and show you some real-life examples. The net result is a log-and-metrics-based monitoring system requiring zero human training or configuration, that recognizes and characterizes novel incidents.