Machine learning for product Recommendations

NineTwoThree has significantly upgraded Cymbiotika's online shopping experience by introducing a machine learning (ML) based recommendation system. This system aims to drive sales by understanding and predicting what customers might buy, based on their similarities with other users.

The goal is to make the shopping experience  more personalized, which is expected to engage  customers better and lead to more sales.

Aiming for at least a 1% increase in sales, this improvement could greatly enhance Cymbiotika's revenue and market position.

Originally, Cymbiotika used a basic system where a customer survey influenced product suggestions. NineTwoThree shifted this approach to a more advanced, behavior-based machine learning system. This new system categorizes customers by their shopping habits, preferences, and responses, leading to more relevant recommendations.

CONCEPT

Cymbiotika, a leading supplement brand, recognized the need to enhance its eCommerce platform's product recommendation system. The goal was to improve their sales metrics, specifically increasing Lifetime Value (LTV) and the First-Time Conversion Rate.
‍
They were relying on an unsophisticated algorithm that used a basic weighted formula for predictive product recommendations. The Cymbiotika team needed a more robust purchasing behavior prediction model to give a better user experience for first-time and returning users.
‍
NineTwoThree proposed conducting a Proof of Concept (POC) to develop a recommendation system aimed at enhancing user retention and increasing customer lifetime value. The primary objective of this POC was to identify the most similar users based on quiz and/or purchase history data to increase LTV for users through tailored recommendations.

Challenge

Cymbiotika's previous product recommendation system was basic and somewhat rigid. It primarily relied on customer responses to a 15-question survey, leading to a generalized product-matching process. This system struggled to adapt to the diverse and evolving preferences of individual customers.

The primary challenge for NineTwoThree was to design a recommendation engine that not only understood customer preferences more deeply but also dynamically adjusted to changing user behaviors and trends.

The aim was to enhance the accuracy of product suggestions, thereby not only boosting the likelihood of first-time purchases but also fostering a sense of personalization and satisfaction among customers, ultimately increasing their lifetime value to Cymbiotika.

As Easy As:

Step 1

Identify most similar users through a survey or historical data

Step 2

Design a recommendation engine that understands changing needs

Step 3

Add A/B testing capabilities to the platform

Step 4

Create a better match between customer preferences and recommendations

solution

In response to this challenge, NineTwoThree implemented an machine learning-based recommendation system on Cymbiotika's Shopify platform. This system diverges from the traditional approach of static product-feature matching. Instead, it analyzes a vast array of customer data points, including past purchasing behavior, browsing history, and demographic information. This holistic approach allows for a nuanced understanding of each customer's unique preferences and shopping habits.

The system was also designed with A/B testing capabilities. This feature enables Cymbiotika to route a portion of its web traffic to the new machine learning system while the remainder continues to use the existing recommendation engine. Such an arrangement allows for a direct comparison of both systems' effectiveness in real-time, providing valuable insights into user preferences and system performance.

impact

The implementation of the new machine learning system is an ongoing process, but early indications show a notable improvement over the previous system. The initial data suggests a more effective match between customer preferences and product recommendations, hinting at the potential for increased sales conversions. Even a modest increase in sales, in line with the 1% target, would significantly enhance Cymbiotika's revenue.

The system's dynamic slider for traffic distribution is a key feature, offering Cymbiotika the flexibility to manage and scale the test as needed. This control is vital for a nuanced analysis and gradual integration of the new system. NineTwoThree's efficient and agile development approach, completing this complex integration within three months, exemplifies our proficiency in deploying sophisticated machine learning solutions for eCommerce platforms, particularly those based on Shopify.