The goal for ab testing is to find best variants that serve business goals. The goals tree could be used to find a metric to optimize. For example if the goal is to increase revenue then revenue per visitor or number of members should be increased. To increase revenue per visitor you could increase average order value or conversion. You go on lower and lower on the goal tree and stop anywhere you want. Ideally you choose the most unoptimized metric metric on each testing

It’s useful to ask following questions when looking for new ideas for to test: What are the most highly trafficked pages? What are the highest value pages? What are the areas we already know that users have difficulties with?

If you have any problem statements use them to generate a new hypothesis. Having problem statement is one of the key elements to generate a powerful hypothesis

Data insights (quantitative or qualitive) also make a good foundation for the experiment. For example you might find that conversion on the sign up form is less then expected. Or you may find that users who did certain action have higher chance of converting.