What advances are possible to generate efficient experimental designs, especially in online contexts? • How should one balance causality and prediction?  

Which matters most and when?  

What strategies exist to combine AB testing and learning in order to trade off the value of current responses with the value from longer term !earnings about consumer behavior?