Amazon, A/B Testing, Cost of Experimentation

I told everyone, “Be afraid of our customers, because those are the folks who have the money. Our competitors are never going to send us money.”

via HBR (2007)

“What you really want to do companywide is maximize the number of experiments you can do per given unit of time. If something’s really big—like the big bet we’ve made on Amazon Web Services—then sure, you can do only a limited number of those, so you spend more time thinking about them and talking them through. Somebody wears the black hat and makes the case for why not to do it, and somebody else puts on the white hat and says why it is actually a good thing to do. But since the outcomes of all these things are uncertain, if you can figure out how to conduct an experiment, you can make more bets. So the key, really, is reducing the cost of the experiments. We have a group called Web Lab that is constantly experimenting with the user interface on the website, getting statistical data from real usage patterns about which interfaces work best. That is a huge laboratory for us, and we’ve put a lot of energy into trying to figure out how to be very low cost with those experiments so that we can run a much larger number of them.

As a practical matter, sometimes it’s very, very hard to dramatically reduce the cost of experimentation. There are areas where conducting an experiment is still, in terms of cost, tantamount to just doing it. But you should always be trying to do that. You should be making guesses and then finding out at lowest cost whether or not they are needle movers.”

—Jeff Bezos, HBR (2007)

Located HBR reference pointed to from reference over here.