Projects at Rakuten #anchor

Ichiba's smartphone search, one of the highest revenue drivers in Rakuten at the time, had an auto loading interface which loaded 100 search results at the time. Each search result included at least 3 separate links:

• link to the item (image and title)
• link to the shop
• link to the review overview page

Some search results also included a link to Ichiba's product comparison sites as a 4th link.

This brought about a set of issues I wanted to solve. Some slower android phones froze when using the old Android browser and trying to load all search results. When navigating to a search result and then back to the search, customers always had to start again at the top of a search page. This meant, even if they viewed result number 72, they started again at result 1 upon returning.

Touch targets were very small (around 15px instead of the recommended 40px), as previous designs were not commonly tested on mobile devices.

After studying customer behaviour through data from Adobe Site Catalyst and lab sessions, in which I could study customer usage of the search pages, I created several possible designs, which went through numerous iterations and stake holder meetings. We settled on a plan which included 2 A/B tests.

The first test was to replace the auto loading with a pagination interface and to find the ideal number of search results per page. The second one would test a new interface.

The first A/B test was a full success, with the winning variant improving

• revenue: +2.3%
• Conversion rate (CVR): +0.07%
• and reduced both the bounce rate: -7.89%
• as well as the exit rate -4.47%.

For the first test I created all designs and coded the HTML + CSS. For the second test I worked with a graphic designer on the new interface and coded HTML + CSS.

For the second test, I created 2 main views (list and grid view) as well as new designs for all search refinement areas and filters. The design language was strictly following Rakuten's styleguide.

While the second test began promising, it ultimately failed. Revene, CVR and exit rates stayed flat, while bounce rates increased drastically. I noticed my mistake on day 2 of a 14 day test, but was not allowed to stop or pause the test to fix the issue.

My mistake was the image size. As part of my new design I had sought to increase the physical by 1/3 and make them into "retina" images (scaling the image down by a factor of 2). In my tests this did not lead to any issues, as I am always optimising my image assets. I also gave trainings on how to do this to different teams in Rakuten.

In real life, images were not well optimised or compressed. Even a 250x250px image could still have a size of over 2 MB per image. Since images were now much larger then before, they were also much heavier, leading to higher bounce rates. Since revenue and CVR stayed flat, I advocated for a new test (with images scaled with a factor of 1.3 or 1.5). It was decided in the end, that another team member would take over and so only my pagination work was fully released.