Increase e-commerce jewelry sales by cross-selling complementary products

Summary

We achieved a 9% uplift in ARPU (4% CR uplift and 9.5% AOV uplift) by introducing a cross-sell carousel in the shopping cart.

Overview

Moonmagic.com is a fast-growing online jewelry store with 430K sessions per month (according to Similarweb). It sells rings, necklaces, earrings and other products in this category.

Our task was to increase the revenue per user (ARPU) metric by increasing the conversion rate (CR) and average order value (AOV).

Research

We looked at the revenue breakdown to identify the key revenue drivers. We found that approximately 25% of shoppers purchase more than one product per transaction. We concluded that users have a significant propensity to buy more than one item per transaction.

So the idea at this point was to encourage the purchase of multiple items in each transaction.

As a result of heuristic UX analysis, we found real estate in the shopping cart to test a cross-sell carousel.

Next, we needed a recommendation engine to use in our cross-sell carousel. A data-driven recommendation engine would work best for this purpose, so we decided to build one. We created a regression matrix to identify different combinations of products that were often purchased together:

We found that there are over 25 combinations of products that are often purchased together.

Including these combinations in our experiment would greatly complicate it. In addition, the role of an A/B test is to validate one hypothesis at a time. Adding the carousel to the cart was our primary hypothesis, so we decided to use the recommendation engine already built into the site. There were two built-in recommendation engines working on the site. We built a regression model to determine which one drove more sales. We found that users who interacted with one of them converted 3.6 times better than the site average. This was our winner.

Hypothesis

We hypothesized that by introducing a cross-sell carousel into the cart, we would increase the average number of items purchased per transaction and increase AOV.

The risk we ran with this hypothesis is that by distracting users from converting in the cart and leading them to other PDPs, we may negatively impact CR. However, even if this were the case, we expected that the increase in AOV would compensate for the decrease in CR.

So, our solution was to introduce a cross-sell carousel in the cart.

Validation of the hypothesis

To validate this hypothesis, we conducted an experiment to measure the impact of the in-cart cross-sell carousel on CR, AOV and ultimately ARPU.

The experiment had the following characteristics

Experiment Type: A/B Test

Traffic split: 50:50

Key metric: ARPU

Targeting condition: We targeted the test only to users who added a product to the cart and viewed the cart. This was done to minimize data noise and achieve statistical significance of the result faster.

Number of users participating in the experiment: 2,518 users

Number of conversions tracked during the experiment: 694 conversions