Case study

WeCook

85 %
Accuracy in detecting customers likely to churn
1 month
To complete data ingestion from multiple sources
More
Advanced customer insights to improve marketing operations

FIND OUT HOW WE DID IT

Having a good data warehouse and implementing artificial intelligence (AI) really solves business issues. We were all surprised by the speed of deployment and the impact on our numbers. Thanks to adviso’s expertise, we were able to deploy the solution in record time. My advice to CTOs and CMOs, and even CEOs: Collaborate more than you ever have before. The future growth of businesses depends on AI… and no AI without clean data!
Jean-Sébastien Crevier
VP marketing, WeCook
1

A PREDICTIVE MODEL FOR REDUCING CUSTOMER CHURN

WeCook, a leader in the home delivery of ready-to-eat meals, wanted to reduce customer attrition and increase customer lifetime value. To achieve this, the company looked for a partner that could build a data warehouse and create a predictive model. The goal: to identify customers likely to cancel their subscriptions and effectively re-engage them.

Adviso designed and deployed a robust data architecture for WeCook on Google Cloud, leveraging platforms such as Fivetran to integrate data from multiple sources: digital tools, media, operational data (invoices, orders, delivery locations, etc.), customer service and payment processing platforms. A predictive model was then trained to identify customers at risk of churn with 85% accuracy.

The success of this project relied on the data-driven culture at WeCook. For 5 years, the company collected its own well-structured data, and essential asset for training AI models and obtaining reliable results. In the future, collaboration between IT and marketing will become increasingly crucial to ensure data warehouses are optimized and capable of generating exploitable insights for marketing.

2

CHALLENGES RELATED TO THE PROJECT

Build a customer data warehouse with several terabytes of storage and develop a model that could predict churn in less than six months—from scratch.
Limit the costs associated with processing data in the cloud. Google DataFusion was initially considered, but did not respect the budgetary constraints.
Process data from multiple sources: Google Analytics 4, Klaviyo, media (Google Ads, Meta, TikTok, Reddit), operations (invoices, orders, dispatch locations, discounts, reviews, emailed ingredients, promotions), customer service (Zendesk), and payment processing (Stripe).
3

OUR METHODOLOGY

Analyze the existing architecture and recommend an optimal solution.
Design a solution based on Google Cloud and data integration platforms such as Fivetran.
Migrate WeCook's customer and operational data to BigQuery using the Google BigQuery native connector for GA4,Google BigQuery Data Transfer, Fivetran, and customized API calls to load WeCook’s customerand operational data.
Set up a medallion architecture to structure the data and provide a 360° customer view.
Develop the churn predictive model by integrating multiple data sources: web events, customer orders, meals, reviews, lifespan, ingredients, promotions, discounts, and even more.