Fresha operates one of the largest beauty and wellness marketplaces in the world, handling 600,000+ bookings per day across 120+ countries. As the business scaled, our legacy analytics stack hit predictable limits: fragmented pipelines, slow refresh cycles and a constant trade-off between data freshness and query performance. This made it increasingly difficult to provide real-time visibility for product, operations and finance teams.

Over the past year, we rebuilt our analytics platform around real-time principles: streaming-first ingestion, CDC-aware processing, and cloud-efficient storage. This session walks through the architecture and the key decisions behind it, following the data path from PostgreSQL CDC (Debezium) through Kafka and Flink, into a lakehouse, and finally into a fast analytical query layer with StarRocks.

Rather than presenting a reference architecture, the talk focuses on what actually worked in production, what failed, and which trade-offs mattered most when unifying CDC streams, historical lakehouse data, and sub-second dashboards behind a single SQL surface.