BlaBlaCar’s Guide to Cultivating a Successful Data Mesh

Data mesh, a hot topic within the data management industry, has drawn the attention of numerous enterprises looking to make data reliable and accessible within a self-service environment. Though the desire to incorporate a data mesh is a good start, the journey of challenges and opportunities to navigate is a winding path that requires enterprise-wide adoption and informed usage.

BlaBlaCar, the community-based travel network for ridesharing, has been guiding their data mesh journey for over a year—and doing so with non-traditional methodologies. Discussing their data mesh strategies in DBTA’s webinar, “The Road to Data Mesh: BlaBlaCar's Journey to Self-Serve Data Analytics,” experts at BlaBlaCar, moderated by Neil Gleeson, customer success at Monte Carlo, explained how they crafted their data mesh to be conducive to both reliability and domain-oriented data ownership, while focusing on their organizational use cases and knowledge needs.

As, “the go-to marketplace for shared travel,” BlaBlaCar maintains 100 million members with 25 million travelers per quarter in 22 countries; between data sources, tooling, data science, lakehouses, and data observability, data is both critical and vulnerable to complexity.

Kineret Kimhi, data analytics engineering manager at BlaBlaCar, offered insight on how BlaBlaCar’s data is entirely crucial toward its organizational success.

“Understanding the huge volumes of data that our marketplace product brings, it’s pretty clear that we cannot move forward and improve our main KPIs without relying on data,” said Kimhi. “We don’t make any decision without using data.”

BlaBlaCar’s data team mission is to deliver dependable data and algorithms to the company, explained Emmanuel Martin-Chave, VP of data at BlaBlaCar. Due to their constant influx of critical data, BlaBlaCar believes that data-informed decisions lead to the most long-term value; creating alignment and autonomy between data and the company, then, is even more critical. 

Building a great data mesh to drive data alignment and autonomy requires synergy between both technologies and people, commented Gleeson.

Tushar Bhasin, senior data engineer at BlaBlaCar, expanded on the specific technology stack that BlaBlaCar employs:

  • Entirely hosted on GCP
  • Uses production databases like MariaDB and Apache Cassandra for data sources
  • Actively uses streaming data, centered around Kafka and in-house tools
  • Consumes a large amount from external data sources like Facebook, Google Ads, and Salesforce
  • ETL is managed with Rivery for quick ingestion and Apache Airflow
  • Uses Tableau and Data Studio for visualization
  • Equipped with Monte Carlo’s data observability platform
  • Leverages MLflow and Kubernetes with Python language

BlaBlaCar’s technology stack lends itself to an effective data mesh, which the enterprise looked to when Martin-Chave began reading material on data mesh and its operation—where it “clicked” for Martin-Chave that the data mesh could address the data problems BlaBlaCar had been facing.

To migrate to data mesh, ultimately, the speakers pointed to three main takeaways that direct data meshes toward success.

BlaBlaCar’s experts explained that first, it’s a mindset change; adoption of the data mesh is as necessary as its operational value. Specifically, convincing and engaging data engineers that owned the data warehouse posed a difficulty for BlaBlaCar, as those teams had to reckon with the most dramatic changes.

It boils down to change management; if an organization has a thorough and effective change management process, the adoption of the data mesh becomes much smoother. Of course, Kimhi pointed out, it still is a technological or architectural change, but primarily, the mindset part must be focused on to drive success.

The speakers next pointed to “stealing” from engineering—leveraging their practices to guide migration. Stealing from monolith to SOA, copying both domain and setups within data engineering teams, aided in BlaBlaCar’s data mesh strategy by making the parallel between software engineer and data management.

“Data mesh is essentially stealing from data engineering,” said Kimhi. “Now we have a much clearer and stronger lineage between data and engineering teams—copy the way your engineering team is built.” 

Finally, BlaBlaCar’s experts emphasized onboarding your stakeholders to prevent negative kickback during the data mesh transition. During this ongoing transformation, stakeholders will become impatient and inevitably have a poorer experience; onboarding them on the problem, the speakers explained, was immensely effective in remediating data mesh agitation symptoms.

The stakeholders are “the people that need to be convinced the most about reorganizing into a data mesh,” explained Kimhi.

For an in-depth discussion about BlaBlaCar’s data mesh journey, you can view an archived version of the webinar here.