key takeaways: 1) judge split test (or new features?) strictly by top 5 metrics of customer behavior: "macro metrics" 2) have engineers only work on the top f

key takeaways: 1) judge split test (or new features?) strictly by top 5 metrics of customer behavior: "macro metrics" 2) have engineers only work on the top feature in the queue, and not put it down til it's done; managers may rearrange queue as often as they desire 3) 5 why's: invest improvement time proportional to size of problem, at each of 5 points of "failure" 4) product feedback is about your CUSTOMER, not about you. Question: Why does the behavior of current customers necessarily predict the behavior of future customers? I understand why it would in the case of IMVU (people aren't paying monthly/one-time fees; they pay based on consumption).. and some other business.. but not, say, for the quintessential web2.0 web app. Can someone please clarify? Don't get me wrong, I think these are killer insights :)