[ajju]

The author is using machine learning and multivariate testing as straw men.

No matter how much multivariate testing you do to optimize your "copy, size, color, font and placement of data fields; the format and text of buttons; the flow of pages", in order to get millions of users to sign up, you still have to a) Get millions of users to your web page b) At least appear to offer something they think may be valuable. Millions of people won't sign up for free TPS reports no matter how well you design your site.

What's more, if the number of "accounts" was the metric everyone used to judge companies, MySpace would be hot hot hot. Active accounts have to be considered and their definition obviously matters.

The author's last argument is the only one that I find to be strong: First spend time building a great product and then focus on optimizing things like font size using multivariate testing. When you build something people want, any traction you get is there to stay.