At a glance
A division of a leading eyewear company provides ecommerce sites for independent opticians. To meet consumer expectations, these sites required a personalised customer experience with product recommendations. Despite access to patient information and past purchase histories, effectively implementing a recommendation system was a daunting task.
The division engaged Credera to optimise conversion and find value in the data. The Credera team leveraged Azure Databricks and our data product methodology to train a recommendation system that utilised customer data to personalise eyeglass frame suggestions. Additionally, a machine learning operations (MLOps) platform was developed to operationalise the machine learning (ML) models, enabling multiple, real-time recommendations to personalise the site to the customers’ activity and past data. The top recommendations the system generated were strongly aligned to purchasing habits and it adapted well to the viewing activity as a customer continued to interact with the site.
The successful integration of the ML pipelines paves the way for broader application of data science in improving customer experience, product optimisation, customer conversion, and market share growth.
Lack of personalisation held back conversion.
A leading provider of white-labelled eyewear ecommerce platforms for independent opticians faced a unique challenge. Their system lacked personalisation, 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 utilise this data effectively to implement a recommendation system.
They recognised that providing a personalised 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 personalisation for existing and new customers alike that was both effective and efficient.
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 optimise 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, operationalising 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, personalising the shopping experience for each customer.
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 personalised 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 personalisation, 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, optimising their product offering, and increasing their market share.
Through the use of personalised recommendations, the company is now better equipped to provide customers with glasses that match their unique style, resulting in a more successful shopping experience.