Cymbiotika creates supplements designed to support everyday health. As the business scaled, the team recognized the need to enhance their eCommerce product recommendation system. Specifically, they wanted to increase Lifetime Value (LTV) and First-Time Conversion Rate. Their existing algorithm used a basic weighted formula tied to a 15-question quiz, producing generalized recommendations that couldn't adapt to individual customer behavior.
Cymbiotika brought in NineTwoThree to build a proof of concept for a machine-learning-based recommendation engine that could identify similar users from quiz and purchase history data, with A/B testing built in to prove it out.
Built a custom ML system on Shopify to predict what customers would buy and increase lifetime value.
Added live split testing to compare the new ML model against the existing recommendation engine.
Delivered a working POC on Shopify within 12 weeks, from kickoff to live testing.
Modeled purchase history, browsing behavior, and demographics to personalize recommendations.
Cymbiotika's revenue growth depended on turning first-time visitors into repeat customers, but their existing recommendation system couldn't personalize suggestions at the level needed to drive those conversions.
Cymbiotika's product recommendation system relied on a 15-question survey, leading to a generalized product-matching process. The system struggled to adapt to the diverse and evolving preferences of individual customers, offering the same suggestions regardless of how a customer's behavior changed over time.
The existing algorithm used a basic weighted formula that couldn't learn from purchasing patterns, browsing history, or demographic data. Every recommendation was static. Once a customer completed the quiz, the system had no way to refine its suggestions based on what the customer actually did next.
Cymbiotika had no mechanism to test whether a new recommendation engine would actually perform better than the existing one. Without a way to compare systems side by side, investing in a custom ML model meant committing budget without evidence it would outperform the status quo.
We approached the project as a focused proof of concept on Cymbiotika's existing Shopify platform, building and validating the recommendation model within three months.
We identified the most similar users based on quiz responses and purchase history data. This similarity model forms the foundation of the recommendation engine, enabling tailored product suggestions that increase customer lifetime value.
We built a machine learning recommendation system on Cymbiotika's Shopify instance. Instead of static product-feature matching, the model analyzes purchasing behavior, browsing history, and demographics to predict what each customer is most likely to buy.
1000s
of sales predicted
We added live split testing with a dynamic traffic distribution slider, letting Cymbiotika route a portion of traffic to the new ML system while the rest uses the existing engine. The team can compare effectiveness in real time and scale gradually.
12 weeks
from kickoff to a live POC on Shopify
Early data from the live A/B test shows a more effective match between customer preferences and product recommendations compared to the previous system, indicating the ML model is learning and adapting as intended.
The ML system replaced static product matching with recommendations informed by purchase history, browsing behavior, and demographics, creating a more personalized experience for first-time and returning customers.
The proof of concept went from kickoff to live testing on Shopify in weeks, letting Cymbiotika validate the ML approach against their existing system without committing to a long development cycle.
The dynamic traffic distribution slider gives Cymbiotika the flexibility to manage and scale the A/B test independently, enabling gradual integration of the new system with continuous performance monitoring.
Build a new platform, or add AI capabilities to existing software, we deliver solutions that work in real operational environments.
Meet with founders Andrew Amann & Pavel Kirillov
Custom AI implementation roadmap
No sales pressure
