This is a great read for anyone who has an onboarding process that is more than one step
One thing to point out: Users who added a custom theme converted better, which can mean two things:
- Those users were already planning on using Ghost and changing its themes - they would have been paying customers, and most paying customers have a need to customize
- They experienced an Ah-Ha! moment when seeing the power of customizing Ghost, and then became converting customers.
This article seems to assume the latter without mentioning the former. Even so, they are still doing much better with this assumption. I’m not sure if number 1 is even actionable, or more of something to keep in mind.
It would be difficult to pull out who would have converted regardless of the custom theme but very enlightening on how a single thing can have such a large impact on conversions.
Ya my money is on correlation and not causation. Way too big of a conversion bump. They could verify by split testing the custom theme feature.
Interesting correlation though none the less.
I disagree. I noticed that before, too – when I really decide to go with a hosting platform, the first thing I do is to install a custom theme. Before that decision I do not invest time (and money, for paid themes) into that. So IMHO it is a causation (however an opposite one – first there is a decision to use the hosting and then installing the theme).
The article, I believe, is not that much about this specific causation/correlation, but about a common method of attacking the conversion problem – collect the data what “achievements” users make and look for patterns to increase said conversions.
I’m pretty sure I could figure out the math based on my university statistics course, but I wonder if someone could publish a blog showing steps-by-step formulas how to calculate correlations in the context of conversions (may be with data from GA or Clicky).
Overall, it’s really interesting that making the custom theme process easier DID see benefits, making this analysis worth it - despite possible flaws in the conclusion made with the data.