10 ms·

The maths looks wrong; the distribution of clicks is only relevant if the total number of clicks stays the same. Essentially a carousel gives you a "second-bit

by ig1 13y ago

The maths looks wrong; the distribution of clicks is only relevant if the total number of clicks stays the same. Essentially a carousel gives you a "second-bite" at your customer, i.e. the second image is a chance to convert those customers who haven't already been converted by the first image. Imagine you have two images one which converts at 40% and one which converts at 20% - but each image converts completely different groups; by carouseling the images you could end up with 40%+20% conversion. In practice you won't because of overlaps, drop-off from people who never see the second image, etc. but it's definitely possible to exceed the maximum conversion of an individual image. (obviously you should A/B test for your circumstances)

yummyfajitas 13y ago
Sure, anything could happen. As Shana says: It is totally possible that there are edge cases where we are wrong. While we’ve personally performed split tests between carousels and their elements and always came up with an individual slide (rather than the whole carousel) as the highest revenue generating source, it is possible that there will be some cases where a carousel is better for conversions. To go with your example, assuming each group has 50%, it's still tough. The actual conversion rate of the carousel in your example will be T x 60% + (1-T) x 30%, where T is the fraction of users who actually wait for the second image. Some simple algebra suggests that for this to exceed 40%, you'll need T > 1/3.
- ig1 13y ago
  The problem with your calculation it assumes that people who are likely to get converted by the first image wait around for the second image and get "unconverted". A good intuitive way to think about it is this: if you're only showing the second carousel image to people who the first image failed to convert then there's zero down-side. In practice there are some users who would convert if shown the first image for longer who you'll lose by rotating to another image, but that's what you need to measure to decide if it's worth having a carousel.
  - shanacarp 13y ago
    the downside is they may bounce instead of waring out the time.
  - yummyfajitas 13y ago
    30% is avg of 20% and 40%. Its the conversion rate of a random person who doesn't stick around long enough to see the second image. I don't think we really disagree - you should test the carousel, and cases where it wins will be edge cases usually.
shanacarp 13y ago
This assumes one important problem that sees to not be common - that the person viewing the carousel will stick around long enough to view the next choice (or all the choices) In an ideal world, what you should do instead is parametric optimization of some sort instead of a carousel - so you can hit that other 20%
snowwrestler 13y ago
You can get the same set of bites faster and more effectively by simply placing both images directly onto the page. This is why Amazon Kindle and Apple product pages tend to be so tall. People can scroll a lot faster than they can wait for a carousel to go through.
- adrianhoward 13y ago
  This. I've personally never seen carousels out-perform alternatives like having multiple images on the page, or multiple landing pages with separate messaging, or just dropping multiple options for the one that's most important. I'm sure that there are some cases where they're an effective option - but I've yet to come across one in testing. As Brad Frost once observed "Carousels exist to keep people from beating the shit out of each other in meetings." - https://twitter.com/karenmcgrane/status/316563205294010370 https://twitter.com/karenmcgrane/status/316563205294010370 And sometimes "people" is just the different people in your head who want to talk about ALL THE THINGS - when talking about just the most important one is what you really need to do.
zecho 13y ago
I have A/B tested carousels extensively at my previous job. What I found was that even after a user interacted with the controls, meaning they specifically chose to see subsequent slides, they still clicked the first slide at nearly the same rate as those who never interacted at all (tested with auto slide on and off and variations of the carousels). Intrigued by the finding, I started doing different sorts of the slides (random, most recently added, oldest, etc.) and found no statistical difference. I never had time to figure out why, though, as I took a new position somewhere else. I have a strong suspicion that users see first as most important, though.

by ig1 13y ago

yummyfajitas 13y ago

Sure, anything could happen. As Shana says: It is totally possible that there are edge cases where we are wrong. While we’ve personally performed split tests between carousels and their elements and always came up with an individual slide (rather than the whole carousel) as the highest revenue generating source, it is possible that there will be some cases where a carousel is better for conversions. To go with your example, assuming each group has 50%, it's still tough. The actual conversion rate of the carousel in your example will be T x 60% + (1-T) x 30%, where T is the fraction of users who actually wait for the second image. Some simple algebra suggests that for this to exceed 40%, you'll need T > 1/3.

ig1 13y ago
The problem with your calculation it assumes that people who are likely to get converted by the first image wait around for the second image and get "unconverted". A good intuitive way to think about it is this: if you're only showing the second carousel image to people who the first image failed to convert then there's zero down-side. In practice there are some users who would convert if shown the first image for longer who you'll lose by rotating to another image, but that's what you need to measure to decide if it's worth having a carousel.
- shanacarp 13y ago
  the downside is they may bounce instead of waring out the time.
- yummyfajitas 13y ago
  30% is avg of 20% and 40%. Its the conversion rate of a random person who doesn't stick around long enough to see the second image. I don't think we really disagree - you should test the carousel, and cases where it wins will be edge cases usually.

shanacarp 13y ago

This assumes one important problem that sees to not be common - that the person viewing the carousel will stick around long enough to view the next choice (or all the choices) In an ideal world, what you should do instead is parametric optimization of some sort instead of a carousel - so you can hit that other 20%

snowwrestler 13y ago

You can get the same set of bites faster and more effectively by simply placing both images directly onto the page. This is why Amazon Kindle and Apple product pages tend to be so tall. People can scroll a lot faster than they can wait for a carousel to go through.

adrianhoward 13y ago
This. I've personally never seen carousels out-perform alternatives like having multiple images on the page, or multiple landing pages with separate messaging, or just dropping multiple options for the one that's most important. I'm sure that there are some cases where they're an effective option - but I've yet to come across one in testing. As Brad Frost once observed "Carousels exist to keep people from beating the shit out of each other in meetings." - https://twitter.com/karenmcgrane/status/316563205294010370 https://twitter.com/karenmcgrane/status/316563205294010370 And sometimes "people" is just the different people in your head who want to talk about ALL THE THINGS - when talking about just the most important one is what you really need to do.

zecho 13y ago

I have A/B tested carousels extensively at my previous job. What I found was that even after a user interacted with the controls, meaning they specifically chose to see subsequent slides, they still clicked the first slide at nearly the same rate as those who never interacted at all (tested with auto slide on and off and variations of the carousels). Intrigued by the finding, I started doing different sorts of the slides (random, most recently added, oldest, etc.) and found no statistical difference. I never had time to figure out why, though, as I took a new position somewhere else. I have a strong suspicion that users see first as most important, though.