Firsthand Perspective: How I A/B tested a Database Versioning tool on HN

aaron-larner-profileAaron Larner is a Madison entrepreneur and CTO of The Art Commission. Aaron shares his experience A/B testing his database version tool DevJoist via Hacker News.


This past Monday I shared DevJoist on Hacker News. I posted around 10:30am CT and quickly reached the front page, hovering around spot #8 until about noon. I've summarized some of the data that I collected while DevJoist was on the front page.

The Tagline Test

Before submitting to HN I had set up an A/B test to determine how well different taglines converted to registrations. Here are the conversion results by tagline:

Tagline Visits Pages / Visit Visit Duration Bounce Rate Registrations Conversion Rate
Collaborative Schema Editing 379 1.74 0:01:15 65.17% 1 0.0026
Database Schema Versioning 369 1.92 0:01:16 60.16% 2 0.0054
Database Versioning Done Right 314 1.82 0:01:27 64.33% 1 0.0032
Version Control For Your Database 397 1.76 0:01:22 65.99% 3 0.0076
Version Control For Your Schema 313 1.76 0:01:27 59.11% 2 0.0064

Version Control For Your Database and Version Control For Your Schema were the winners in terms of registrations. It's also interesting to look at the time it took people to register based on which tagline they saw:

Tagline Avg. Time to Register
Collaborative Schema Editing 1.91 minutes
Database Schema Versioning 1.19 minutes
Database Versioning Done Right 2.88 minutes
Version Control For Your Database 0.94 minutes
Version Control For Your Schema 1.87 minutes
Product Tours

Within the app, I added a little product tour. The tour consisted of small popups that highlighted the functionality of different elements within the app. Each column in the table below represents one of the tour popups. They were presented to users sequentially starting with "Save a version."

The numbers in the table below represent unique views of each tour popup

Tagline Start a project Save a version Projects dropdown Diff and merge Collaborate with other developers Don't forget to register
Collaborative Schema Editing 22 6 5 4 3 3
Database Schema Versioning 23 0 0 0 0 0
Database Versioning Done Right 15 6 4 4 4 4
Version Control For Your Database 27 0 0 0 0 0
Version Control For Your Schema 23 4 3 3 3 3

I thought it was interesting to see that almost everyone closed the first tour popup immediately, but the people who continued after the first popup were very likely to continue all the way to the end. 3 of the visitors who continued past the first popup registered. That's a 18.75% conversion rate, which is very similar to the 18.33% conversion rate of users who clicked the call to action button. In the end, it looks like the product tour has little to no effect on final registrations.

How To Run Your Own Tests Using Google Analytics

Here's how I ran my tests. For each new visitor I generated a random integer between 1 and 5 to determine which tagline that user would see. I then stored that tagline in a cookie on the users machine, so that they wouldn't see other taglines if they refreshed the page. When I rendered the page I insterted the appropriate tagline for the user and set a custom variable in google analytics (with their javascript API). I also sprinkled GAN event API calls throughout my javascript to keep track of a slew of other information, such as:

  • When users registered
  • Which tour popup boxes they saw
  • etc.

With a mixture of advanced segments and custom reports within Google Analytics, I was able to come up with all of the data that you see above. See the full blog post on the DevJoist website for more implementation details here.

Keep up to date other DevJoist news. Follow me on twitter: @devjoist

Disclaimer: The purpose of running this analysis was mostly to get practice and get in the habit of using data to drive decision making. I haven't done a rigorous analysis on the statistical significance of the results. Many of them are likely not statistically significant because of the relativly low traffic volume.