This sort of segmentation exists in many, many free trials. Sometimes the causational arrow can be difficult to tease out. For example, one's best customers often engage higher with the app on every conceivable dimension, so if you happen to test a particular feature, you'll see adoption of that feature correlates with conversions and might be inclined to e.g. try to juice adoption via promoting in onboarding. That works only some of the time.
In general, though, strong endorsement of figuring out what people care about quantitatively and qualitatively, instrumenting that, and then testing early in the funnel improvements to it. The gold standard is, of course, not doing retrospective analysis but rather split testing the intervention. I sympathize that this is difficult for a lot of SaaS apps as you need high trial volumes to make it work.
I can't share client results but my blog probably has two or three write ups of 10%+ lifts for BCC doing this.
The title is wrong. Ghost didn't increase their conversion rate by 1000%. They found a segment of their user base that converted 1000% more than another segment.
It is not included in this post whether the overall conversion rate actually improved substantially after the changes were made.
Hey Jason, that's actually pretty true. We found a segment which converted 1,000% better and then managed to increase the number of people in that segment by 370%
The title of the post is "How we Figured Out What Makes People Love Ghost 1,000% More" - which I think is still pretty accurate :)
> The title of the post is "How we Figured Out What Makes People Love Ghost 1,000% More" - which I think is still pretty accurate :)
I disagree. If you choose the bucket "people who subscribed to Ghost Pro", they would convert ∞% better than the bucket "people who didn't sign up for Ghost Pro". That doesn't mean that signing up for Ghost Pro makes people love Ghost ∞% more.
More generally, how can you tell if people installing a custom theme makes them love Ghost more, or if people install a custom theme because they love Ghost more?
Getting more people into the bucket of "people who installed a custom theme" doesn't actually help your bottom-line conversion rate until you can establish unequivocally that the people you've added to the bucket convert at the same rate as the bucket did before they were added to it.
> Getting more people into the bucket of "people who installed a custom theme" doesn't actually help your bottom-line conversion rate until you can establish unequivocally that the people you've added to the bucket convert at the same rate as the bucket did before they were added to it.
This is incorrect. Let's say we have two buckets, A and B, where bucket A is people who haven't used a feature and bucket B is people who have. If some testing shows that people in bucket B convert far better than people in bucket A you make a change to get more people into bucket B. The change is successful if more people end up in bucket B and bucket B still outperforms bucket A and the total contribution coming from bucket B is higher. Bucket B doesn't need to perform equally well before and after the test.
Using the numbers from the link bucket B was performing at 10x the conversion of bucket A. Out of 10,000 users bucket B contributed 70 subscribers. After the changes bucket B received almost five times the users as previously and converted at four times the rate of bucket A. Out of 10,000 users bucket B contributed 884 subscribers. If the change had no impact we'd expect more users in bucket B but not more subscribers. These numbers aren't really accurate since I took them from the diagrams in the article and they change what they're looking at between the two. To be correct you need to compare the same thing before and after changing a feature.
> and then managed to increase the number of people in that segment by 370%
Did this end up raising overall conversion rates proportionally? It could be that users who would already have converted in the end are more likely to bother to set up a custom theme.
A segment which did not exist prior to their changes, thereby increasing their conversion rate from X to 11X, as there was a 10X increase in conversions post-change.
The segment did exist - users who added a custom theme. Of that segment, they converted 10X more than those who didn't add a custom theme. Ghost then responded by making it easier to add a custom theme, but we aren't told what effect that had on overall conversion.
How do you know that you have the cause and effect the right way around? After all, it could well be that users add a theme because they are about to buy the service, and figure that they should spend some time customising their pages now that they have decided to sink money into it.
Ok, so that may not sound totally convincing, but in general I think it is a big problem with the analytics in the article. If you can't tell if A caused B or B caused A, then you can't reliably act on it.
There are "ah ha!" moments, and there are also "oh no!" moments.
I love Ghost's simplicity for blogging, but then I spotted an egregious spelling mistake in one of the articles I'd published. I went back to the editor to check where I'd missed the ubiquitous red squiggly line, and found there wasn't one. No squiggly line, and in fact no spellchecker at all!
Given that spelling is very important for blogs, and that all browsers have very effective spellcheckers built in, I'm at a loss to understand why this wasn't bug number 1 to the developers. They use a markdown editor that somehow kills browser spellchecking, but there are several alternatives out there and this was a definite "Oh no!" moment for me.
This data intuitively makes sense -- the added theme gives users a sense of "this is my product," where before it was "this is a Ghost product that I'm using right now."
As a user, however, I'm annoyed by videos, and hardly ever watch tutorial videos. Why spend a minute and a half watching something when I can read it much faster? Especially if setting up a theme is as easy as you're trying to make it -- shouldn't it be obvious how to do it?
I find it a little suspect that you had exactly 3400 people watch your training video and exactly 6600 not watch the video. Those seem like very convenient statistics. Am I missing something?
Hey Adam, great question! I took the conversion %'s measured and applied them to a base sample size just to make the numbers in the post a little bit more easy to read/understand. Hope this helps clarify
This goes to show how important analytics is in a business. They found something they could market better in their own software by understanding their business more. Big ups to Ghost for that discovery.
In general, though, strong endorsement of figuring out what people care about quantitatively and qualitatively, instrumenting that, and then testing early in the funnel improvements to it. The gold standard is, of course, not doing retrospective analysis but rather split testing the intervention. I sympathize that this is difficult for a lot of SaaS apps as you need high trial volumes to make it work.
I can't share client results but my blog probably has two or three write ups of 10%+ lifts for BCC doing this.