Denial, anger, bargaining, depression, & acceptance: The A/B testing lifecycle. | Hacker News Reader

Denial, anger, bargaining, depression, & acceptance: The A/B testing lifecycle.

benplesser.com

24 pointsbjpless11 comments

Getting emotionally invested in your tests is a good recipe for an emotional rollercoaster. Don't do that.

Instead try a bunch of things, and see what sticks. Be curious, not invested.

That's a great point, echoing the part of this post about not being results-oriented. Process-driven is the way to go!

This is a great piece. I'm doing a lot of testing for advertising right now, and the emotional rollercoaster - no matter how sanely you attempt to approach it - can be pretty intense. It's good to see someone writing about that side of the process!

thanks thenomad. There's a funny disconnect between the rational, detached self and the insane subconscious self present in most people.

What is the most frustrating A/B test that you've run?

Are you A/B testing different Kevin Costner photos right now? Clever. I like it.

Off-topic but what do you guys use to run A/B tests?

We're currently using Google Content Experiments. The Pros are that it's incredibly easy to get a test started thanks to the presence of Google Analytics scripts in the rest of the website. The major Con is the lack of true multivariate testing.

We experimented with Optimizely but had a bad experience. It seemed as though their javascript was synchronous and blocking. We may have been able to customize that but did not dig deeper.

Another flaw in Google Content Experiments (which was mentioned in the article) is that it doesn't distribute traffic evenly. This is a big bummer with seasonal products/services where having the control start getting 80% of the traffic during a high conversion period makes the test useless. Overall, even though testing is easier in some ways, I definitely miss Google Website Optimizer.

For our logged in user A/B testing (often relating to email), we've built an in-house solution that uses a MOD based system to track group performance (as opposed to storing each group member in our databases).

Pretty clever implementation by the data scientist on our team.

We've looked at Optimizely and Google, but we prefer VWO.

Reply on news.ycombinator.com