Hey Lawrence, when you're at work does anyone ever talk to you about "Actionable Metrics?"
No. NO. Shit, no, man. I believe you'd get your ass kicked for saying something like'gat man.
For example, say you're considering a cool new CAPTCHA for your registration page. The article would advise doing an "A/B split test" before phasing out the old CAPTCHA.
Such a test would involve creating an alternate registration page with the new CAPTCHA, and randomly directing x% of users to the alternate page. You could then obtain registration completion % from both groups, which would drive your decision to switch to fully switch to the new CAPTCHA.