Ask HN: How to generate better hypotheses?

There is a great deal of literature published online on testing, tracking, measuring, optimizing, etc. (e.g. customer development, A/B testing), but there are less resources about how to generate the hypotheses to test in the first place. The ability to generate better hypotheses naturally results in less testing and generally more efficient use of resources.

How does one improve one's hypothesis-generation abilities? I have had many ideas in the past that never survived the "customer discovery" stage, simply because whatever I came up with did not seem to strike a chord with my target customers. A/B testing is theoretically good, but it's hard to generate hypotheses about why visitors are not converting.

What is your take on this?