An online retail fashion company has its entire sales operation on desktop and mobile websites. Through merchandising and featuring various promotions it attempts to attract and retain certain user segments and profiles. However, these efforts are not personalized enough to get the level of results the retailer wants.
With Rumble Px the company could effectively offer personalized user experiences, generating bigger cart-sizes and driving up sales. In this specific optimization campaign the company chose to display personalized product bundles offering free shipping and discounts during the checkout process via badges placed on the promoted items.
Rumble Px in Action
The measuring KPI for the campaign was Cart Size, specifically looking to see an increase in average cart size at checkout.
Although Rumble Px offers a variety of business tools to immediately start with, our client wanted to design its own UX/UI solution and apply it using our modular platform. The customer’s product team applied its own design solution, letting the Rumble Px platform do the rest.
The optimization drivers for the campaign are badges, a UX element added to specific items. The retailer created several types of badges such as free shipping, 10% discount and coupon for next purchase. The Rumble Px platform, with its personalization and machine learning capabilities, offered the best badge to the right audience, constantly measuring results and adapting the offering versions accordingly.
The retailer knew their customers well, and therefore felt confident to define the target segments for the optimization manually, setting it to 10 distinct segments revolving around demographic data and top categories attributes (for example, young females buying sneakers).
In this case, our algorithm needed to answer two main questions:
1. Which badge should be displayed for which segment?
2. Which items should be decorated with a badge?
For a short period of time, we used all types of badges across all segments on randomly selected items. We then made sure that every user receives a single type of badge – this helped us answer the first question.
We then applied the Naive-Bayes model on the “AddToCart” event count per user-segment.
Then for a short time, we ran another experiment where we used our knowledge from the previous trained model, and presented a single badge type for each user-segment that were randomly scattered on items.
Using the data from the second experiment, we could train a ‘Neural Network’ model to decide which item was optimal to have a badge per segment.
This is a snapshot of the training process using AzureMl.
Once we trained a Neural Network model, we could generate predictions for each item, with a live stream of data regarding the trendiness of the item.
In applying Rumble Px to the Retailer in a production environment we left 10% of the users as a control group, which meant they would receive badges randomly on items. The control group not only served as a performance benchmark, but also allowed us to detect if the optimization group had a specific instance of lower performance, we would be able to react by retraining the model based on that data.
Within just days of running the campaign, Cart Size showed a significant increase with an average of 11%. After 3 weeks the average went up even higher to a 14% increase, thanks to the machine learning improving the model and offers.