How Brevo accelerated model development by 80% using ZenML

• Brevo, formerly known as Sendinblue, leads in leveraging machine learning to enhance email marketing services and streamline operations.
• To refine their offerings and address scaling challenges, they integrated ZenML to optimize their ML integrations.
• Post-integration, their ML deployment time has decreased by 80%.
• They've achieved significant outcomes: 5 models in production, improved fraud targeting, heightened customer satisfaction, and reduced workload.
• They've enhanced team productivity and enabled end-to-end ownership.

The Challenge: Overcoming Scalability and Efficiency Hurdles

At the beginning of their MLOps adoption journey, Brevo's ML landscape was fraught with challenges that hampered its ability to fully leverage the power of machine learning:

• Limited Monitoring and Impact Visibility
  ‍Hindered performance insights and customer impact assessment due to absence of effective monitoring.
• Isolated Development and Deployment
  Lack of cohesion in deployment strategies due ML models operated in silos.
• Slow Development Cycle
  Delayed business value realization due to month-long development-to-deployment process.

The ZenML Advantage

Brevo's transition to ZenML marked a pivotal shift in how they approached machine learning. ZenML offered a suite of features that directly addressed Brevo's challenges.

Homogeneity of Infrastructure & Centralization

Brevo deployed models using various tools within Google Cloud Platform (GCP): Vertex AI, BigQuery ML, and potentially Cloud Function. Despite their unique features, ZenML unified these tools for data scientists, maintaining agility and speed.

ZenML provided a unified experience layer for the data scientists to leverage each individual service, without losing their agility and speed.

Incremental deployment

ZenML provided the flexibility to begin with a compact pipeline operating on a local scale, ensuring detailed inspection of every phase up to a full-scale production pipeline scheduled on the cloud. This incremental approach offers a multitude of possibilities, encouraging experimentation, testing, and evolution of the pipeline.

Reliable ML models that are reproducible and auditable

Adopting ZenML ensures reliable, reproducible, and auditable ML models with strong data and model versioning, guaranteeing robustness, efficiency, and full traceability. Brevo gained clear insights into model performance, data usage, and experiment outcomes, reducing troubleshooting time with complete model lineage for review.

The Results: A Leap in Operational Efficiency and Market Reach

Brevo's adoption of ZenML yielded significant improvements across their ML operations and business outcomes:

• Increased Model Efficiency
  ‍With 5 models now in production, Brevo improved fraud targeting, boosted customer satisfaction and reduced workload.

• Automated Delivery & Faster Time-to-Market
  ‍ZenML drastically reduced ML lifecycles from months to days with its automation capabilities, enabling rapid response to market needs.

• Enhanced Team Productivity
  ‍The collaborative features and automation provided by ZenML have streamlined the workflow, allowing the team to focus more on innovation rather than operational overheads.
• End-to-end ownership
  ‍A team of 3 data scientists can independently and autonomously deploy a new ML use-case in a matter of days, all the way from research to production.

The Business Impact: Transforming Email Marketing with AI

ZenML not only optimized Brevo's ML operations but also revolutionized their business:

• Improved Customer Engagement
  ‍ML-driven insights boosted email marketing effectiveness.
• Operational Efficiency
  ‍Automation reduced time and resource requirements, optimizing resource allocation.
• Market Competitiveness
  ‍ZenML's rapid deployment capabilities positioned Brevo as an email marketing leader.