> Part of the reason why many companies jumped on this bandwagon is also because 'customizability' is hard(er) to build in, and certainly more expensive to maintain.
It's also harder to A-B test with so many variables. If tests aren't statistics significant, the value of user analytics and UX experimentation decreases from a "% lift" perspective. It's harder to know if a feature change or a user defined config had a causal relationship to some other metric.
It may be the A-B testing tail wagging the dog.