The Challenge

Lack of personalization held back conversion.

A leading provider of white-labeled eyewear ecommerce platforms for independent opticians faced a unique challenge. Their system lacked personalization, making it challenging for customers to identify the eyeglasses that best suited their individual style. Despite having access to detailed patient information from the opticians and a comprehensive past purchase history for each customer, they were uncertain how to utilize this data effectively to implement a recommendation system.

They recognized that providing a personalized experience would enhance the shopping journey, but the challenge lay in creating an adaptable system that harnessed their available data to provide accurate recommendations for each unique customer. They needed a solution that would integrate seamlessly into their existing system, offering personalization for existing and new customers alike that was both effective and efficient.

The Solution

Creating a machine learning recommendation system that learns dynamically.

To address this challenge, an ML pipeline was developed using Credera’s data product methodology and leveraging the Azure Databricks Unified Analytics platform. The pipeline was designed to train a recommendation system, aiming to optimize the ecommerce experience by suggesting eyeglass frames based on each customer's past purchase history, demographic details, current viewing activity and product margin information.

The solution went beyond just training the model. An MLOps platform was established, operationalizing the ML models into a live environment. The platform featured a batch prediction service developed over Databricks Delta to generate content-based recommendations. This service was integrated with an Azure Function API and a Redis-based online feature store. The resulting ecosystem was able to serve multiple real-time recommendations, personalizing the shopping experience for each customer.

The Results

Enabling increased recommendation-driven sales.

The outcome of this integration was highly promising. During the validation testing, the recommendation system demonstrated a high level of accuracy in understanding customer preferences, with a significant percentage of past actual purchases included in the top five frame recommendations generated by the system. Additionally, when such data was unavailable, the system was still able to adapt to each customer, providing recommendations based on the eyeglass frames they were viewing and what was popular at their store. This meant a highly personalized experience for anyone visiting the site, enabling a higher conversion rate and the ability to put more first-party eyeglass frames in front of interested buyers.

To further validate the effectiveness of the recommendation engine, plans were laid out for an A/B test that would measure any changes in product margin and customer satisfaction following the introduction of the recommendation engine. Initial hypotheses predicted a 10% lift in both areas, anticipating an increase in customer satisfaction due to improved personalization, and subsequently, a rise in profit margin.

Looking ahead, the MLOps platform paves the way for the company to expand its use of data science capabilities rapidly, making them well positioned to continue improving their customer experience, optimizing their product offering, and increasing their market share.

Through the use of personalized recommendations, the company is now better equipped to provide customers with glasses that match their unique style, resulting in a more successful shopping experience.