Determining the best test to run depends on your website and its goals. For example, a blog owner may want to increase time on site or page depth, whereas a consulting company may want to increase downloads of its industry report.

For your initial tests, start with goals such as clicks or pageviews over metrics such as time on page or percent of page scrolled. Experiments affecting the latter need to be controlled rigorously, and intent is often difficult to understand.

For example, if the average visitor's time on Page X increases 30 seconds, is that necessarily a good thing? What if your Control Page was more efficient at delivering information, and people find the Treatment more confusing, and thus spend more time trying to make sense of it? Did your bounce rate increase?

So, start with the low-hanging fruit of action-based A/B testing.