[bhaavan]

"Getting out of the building" in metmorphical send might give you view from people / customers you can visit. This is likely to be a biased sample, and unrepresentative of the true customer.

Running A/B tests is IMO, a good proxy, to understand customers unbiased binary views on feature changes, and objectively better.

[madeofpalk]

Talking to users lets you know what the problems are.

A/B tests let you know if you've solved the problem.

[DamilolaA]

True!

[saberdancer]

A/B tests work only for features you've built. You don't get insight into what customers say they want, you only can observe if they like what you built or not.

There is a use case for both.