5 ms·

This really reminds me of the depressing way that Google optimised the blue colour of a button, while ignoring all other considerations: http://stopdesign.com/

This really reminds me of the depressing way that Google optimised the blue colour of a button, while ignoring all other considerations: http://stopdesign.com/archive/2009/03/20/goodbye-google.html http://stopdesign.com/archive/2009/03/20/goodbye-google.html If you trust your metrics and nothing else, you have to be very sure that your metrics encompass every aspect of the reality you are modelling. If they just tell you about clicks and sales, they might be missing longer-term objectives like user satisfaction and retention.

harryf 13y ago
Meanwhile another interesting read - http://jeffdechambeau.com/friending-fast-and-slow.html http://jeffdechambeau.com/friending-fast-and-slow.html
- grey-area 13y ago
  Thanks, that's really interesting.
gfodor 13y ago
I've never worked at Google and I can respect the sentiment, but it seems to me that if you are a skilled designer that Google's massive audience and experimental capabilities can cut both ways. It means that you have to prove even the most minor design changes, but it also means that you have the ability to really prove your work is helpful and not fool yourself. If you make a few small wins you surely can build up the credibility to take bigger steps -- the catch is though in my experience bigger steps never improve metrics anyway and are a waste of time since you end up having to throw more work out. I would kill to be able to run a 0.1% test on a large change and not disrupt things too much, while still getting statistical significance. Few places other than Google seem to have the traffic to do that.
- grey-area 13y ago
  I'd dispute that many design changes can be proven in any meaningful way. There are too many variables, and optimising for clicks on one button might affect all others on a page. It's very hard to measure everything at once, and it doesn't sound like Google tried in this instance.

harryf 13y ago

Meanwhile another interesting read - http://jeffdechambeau.com/friending-fast-and-slow.html http://jeffdechambeau.com/friending-fast-and-slow.html

grey-area 13y ago
Thanks, that's really interesting.

gfodor 13y ago

I've never worked at Google and I can respect the sentiment, but it seems to me that if you are a skilled designer that Google's massive audience and experimental capabilities can cut both ways. It means that you have to prove even the most minor design changes, but it also means that you have the ability to really prove your work is helpful and not fool yourself. If you make a few small wins you surely can build up the credibility to take bigger steps -- the catch is though in my experience bigger steps never improve metrics anyway and are a waste of time since you end up having to throw more work out. I would kill to be able to run a 0.1% test on a large change and not disrupt things too much, while still getting statistical significance. Few places other than Google seem to have the traffic to do that.

grey-area 13y ago
I'd dispute that many design changes can be proven in any meaningful way. There are too many variables, and optimising for clicks on one button might affect all others on a page. It's very hard to measure everything at once, and it doesn't sound like Google tried in this instance.