undefined | Better HN

0 pointspatio1116y ago0 comments

That is a failure of the imagination. They certainly could -- whether it is worth doing or not is another question, but hey, that is what God gave us A/B testing to figure out.

"Hey user, it looks like you came to us today from Google searching for [Facebook login]. Did you know that there is a better way? Type facebook.com into [blah blah blah]. Try it now and we'll give you 5 free credits for [without loss of generality: FarmVille]!"

"Great job! You should do that every time. If you do that to log into Facebook the next five days you use the service, we'll award you a Facebook Diploma and give you another 10 free credits for [without loss of generality: FarmVille]!"

On the back end, you show the above prompts to N% of your users who you detect coming to the login page from Google search results (this is trivial -- check the referer). You then compare any user metric you want for the "Was Shown Facebook Login Course" population and "Complete Facebook Login Course" population with the population at large. Kill the test if it hurts your metrics, deploy it sitewide if it helps them.

0 comments

2 comments · 2 top-level

patio11OP16y ago

Incidentally, I was talking to a fellow Japan-based HNer the other day, and we both independently talked about IMVU. You know why IMVU is the coolest technology company I've ever heard of even though they make a product which I couldn't care less about? Because if this happened to IMVU, the most junior engineer in the company could code up the above behavior and deploy it live in less time than it took to read the blog post describing the problem, in full confidence that the production site would not break.

The possibilities of that development model are staggering. (Want to know how you can run rings around more established competitors? If you're learning at that speed and they're learning at the "release every six months" speed you will bury them.)

pyre16y ago

Well played.