Making the LinkedIn experimentation engine 20x faster

Interesting article on the technical aspects of scaling such a service, no doubt. One question remains - was any of this necessary? It's amazing how much work their machines are doing to analyze the data, and it sounds like all decisions are made on data, but I would love to see is an article that defends all this data in exchange of humans making decisions. Is this algorithmic approach optimal for the outcome, or is it optimal for decision making? Would love for someone from LIN to explain the big picture!

> It serves about 35,000 concurrently running A/B experiments,

With that amount of optimizations, you would expect them to have the best user experience in the world. Do they? I think there is something severely wrong with the current A/B testing methodology.