How Lights Online increased CVR |READ NOW
Small Changes for Big Impacts on Conversion
Siddharth Dalal
June 16, 2022

At Skafos we make interactive personalization tools for eCommerce. Our most popular one is an interactive product recommender. Basically just like the regular recommendations you see on eCommerce pages but more interactive.

We measure the effectiveness of our recommendations in several ways, the most important being how many people actually click through to a product page. We also measure user engagement, something that is not really even a thing for most recommenders.

On a Skafos Recommender, you can instantly “Like” or “Dislike” recommendations, and those are immediately factored in to the rest of the recommendations. So we have several user engagement metrics that we measure including:

  1. Click Through to Product Page
  2. Like / Dislike Click
  3. Scroll

Everyone knows mobile usage is now the majority of eCommerce. As an example, at one of our larger customers, 82% of the users are on mobile. We also discovered that, no matter how flashy we make our buttons, users prefer to scroll as the most used mechanism to view additional products.

So we decided to make our scroll experience the best in class and our team put on their thinking caps. We came up with two goals

  1. Our scrolling must be smooth no matter what, especially more so on mobile
  2. Recommendations should magically get better as they scroll if we have collected other data

Here’s a picture that speaks a thousand words about what happened when we rolled out our changes. The vertical red line is the day the changes were rolled out:

You can see that not only did our click-through rate increase significantly, but it also became a lot more consistent.

So what exactly did we do? I’m giving away our secret sauce here.

  1. We preloaded a lot more products. And when a user was halfway through the list, we loaded more dynamically. This way scrolling was smooth no matter what and product loads kept going until we ran out of products to show.
  2. Normally, when a user clicks like, we immediately load in new products into the recommender based on all their cumulative likes. However, that didn’t change the products that were already loaded. Now we also swap out all the products in the recommender that the user hasn’t seen yet. So if a user has interacted with the recommender, all future products they see when they scroll are more relevant.

Now that we natively support A/B testing, we welcome you to pit your existing recommendation system against ours and see the difference.