Machine learning allows e-commerce businesses to create a more personalized customer experience. With the right use of machine learning businesses can achieve higher customer satisfaction, lower cart abandonment rates and get better sales results.
FreshDirect asked us to introduce an environment where their collected data can be utilized for various business cases and where machine learning algorithms can be deployed and tested. Since they knew that experimenting with ML always has high-risk involved, they aimed to have a vendor-free environment where cost can be saved and their data is as secure as possible.
Business Analysis - Project Management - Machine Learning Services - DevOps - Web Development - QA
Since avoiding vendor lock was one of the key requirements for the project we decided to use Apache PredictionIO - an open-source machine learning server - to build the services on top of. Although the ML infrastructure is hosted on Microsoft Azure cloud which provides scalability and helps in the need of future maintenance, the infrastructure can be moved to any cloud provider in the future with the help of the DevOps tools (Docker) we utilized.
The first service we introduced on the infrastructure is personalized product recommendations: recommendations based on user behavior, recommendations based on product similarities, recommendations at checkout based on products in the cart, recommendations based on product popularity. All these recommenders got introduced into the existing customer-facing store and they all wired into the analytics tools used company-wise.
After fine-tuning the initial algorithms to reach the set business goals a new experiment began. A new predictive recommendation was implemented with a goal to discover the customers’ purchase habits and recommend products for them based on this knowledge. The solution tries to predict items a customer probably needs in her next order.
Apache PredictionIO, Python, Microsoft Azure
