A recommendation engine in just 10 days

Background

ManoMano is a French marketplace specializing in DIY and gardening products. They wanted to increase the average number of items per basket. We detected similarities between catalog photos using a convolutional neural network provided by the open-source OpenCV library.

The Challenge

We created a recommendation engine that increased the conversion rate by 1.7% in 2 weeks.

To implement it quickly for 10 product categories, we used a simple and effective solution: a neural network called VGG16, which is already available and pre-trained.

The Solution

We developed recommendation engines, including one that uses image recognition to suggest products similar to those searched for by visitors.

To ensure the quality of the recommendations in the face of increasing sales, we monitored customer service complaints. We cleaned this data and integrated it into a database accessible to analysts and a Business Intelligence dashboard connected to our datalake.

Results