top of page

Abandon Basket Recipe With Thalys


User Story


“As a CRM Manager I want to retarget visitors who abandoned their cart via email and push notifications to push them to finish their checkout and therefore improve the conversion rate and revenue”


About Thalys


Thalys is a high-speed train service connecting major cities in France, Belgium, the Netherlands, and Germany. Thalys.com is the company’s largest sales channel today. 


Despite being unable to make substantial infrastructural changes due to an impending merger with Eurostar, the business pursued ambitious objectives. Although a significant amount of customer data was collected, it was underutilized, the data and technologies existed in silos, hindering integration and there was a strict commitment to respecting customer data privacy, ensuring its protection.


Our mission is to help Thalys identify, prioritize and leverage the existing customer data in order to drive business impact. We use user stories in order to move from idea to production-ready use case. 


Everyone in the field of marketing would agree when we say that this is a fairly simple user story to put in production if it would not be for the specific context Thalys was in as described above. The context makes that changes to the production environment could not be pushed through nor could any owned systems be altered due to code freezes. This means alternative solutions had to be developed in order to continue to drive revenue uplifts considering the limitations put in place. 


Ingredients





Recipe


Step 1: Tracking Customers who did not complete their cart 

The main challenge Thalys was facing was that their email platform responsible for sending emails  had no notion of baskets being abandoned. In theory, and practice, an easy fix would be to deploy the required code from the email platform directly, would it not be for a complete freeze on code freeze.


Therefore an alternative was needed. We leveraged the existing Google Analytics 4 deployment to recognize which visitors were abandoning their cart during the checkout process.


As GA4 doesn’t allow PII data to be ingested, an identity resolution logic needed to be constructed down the data pipeline by matching pseudo identifiers with data residing in the CRM.


Step 2: Ingesting Data to BigQuery

Once captured, we had to transform the data in the required schema. This is where DBT came in handy. In a nutshell, DBT is a data engineering tool. It offers a development framework using modular SQL to transform your data, centralize your code and collaborate with your team on a single source. Here is an article we wrote on the topic. 


Using DBT, we made the required transformation on the raw data and created a user table flagging all visitors who have abandoned their cart and their email address. This table can be seen as an audience ready to be used by their Email Service Provider (Adobe Campaign) & Notification Push Platform (Airship). 


Step 3: Syncing Audiences with downstream destinations

With the customer data stored in BigQuery, the next crucial step was to create audiences and initiate targeted marketing campaigns. To accomplish this, we chose Census as the reverse ETL solution. 


Census enables seamless data synchronization between BigQuery and Adobe Campaign, a leading marketing automation platform. By leveraging Census’ integration capabilities, businesses can connect the audience created with Adobe Campaign.


Step 4 : create the campaign

Every communication initiative can be summarized as an audience, a message and a channel. In this context, we already have addressed the audience and the channels. The last step was to validate the message (tagline + visuals) for both the emailing campaign and the notification push. Once validated, Thalys’ CRM team completed the set-up of the campaign and pushed it live. 


As a final note, and in order to have a safety net, an exclusion rule was added in the Adobe campaign to exclude all recent purchasers. 


Result


Results were striking with more than a 12x return on investments. Considering here that the investment is represented by the sum of Human37’s consulting fee and the licensing cost for the new technology we had to deploy (BigQuery, DBT & Census). Revenue is measured using a combination of transactional database data and attribution leveraging GA4.

Comments


bottom of page