Fetch's FAST BI: ClickHouse Powers Conversational Analytics

Agentic Analytics Meets Real-Time Data

Fetch's successful implementation of FAST on ClickHouse Cloud highlights a compelling convergence of AI-driven analytics and high-performance data warehousing. The core innovation lies in democratizing complex ad-hoc querying through a conversational interface, transforming raw consumer data into actionable insights for non-technical teams. The ability to handle 150 billion rows daily and deliver sub-second query responses for intricate, multi-dimensional questions is a testament to ClickHouse's scalability and speed, particularly its compute isolation feature which prevents workload contention. This architecture effectively bridges the gap between data availability and business agility, allowing teams like sales and commercialization to iterate rapidly on campaign ideas and partner pitches, a significant departure from traditional, time-consuming BI workflows.

However, a notable aspect is Fetch's chosen data loading strategy: a nightly full refresh into wide, versioned tables. While this simplifies LLM interaction by providing a stable query target and easy rollback, it implies a trade-off in data freshness for certain use cases and might not be optimal for scenarios demanding near-instantaneous updates to the entire dataset. The article also doesn't delve deeply into the LLM's internal mechanisms for query generation, semantic mapping, or the specific prompt engineering techniques used, which are critical components of the FAST system. Furthermore, while ClickHouse Cloud's cost-effectiveness is mentioned, a detailed comparison with alternative solutions based on total cost of ownership (TCO) for such a data volume would add significant value. The reliance on Snowflake as a source for the nightly load also introduces an external dependency and potential cost factor not fully explored.

Key Points

• Fetch built FAST, a conversational BI agent, on ClickHouse Cloud to handle over 150 billion rows of data daily.
• ClickHouse Cloud was chosen for its fast ad hoc query performance, compute isolation, and simplified batch loading via ClickPipes.
• FAST enables non-technical users to query data in plain language, reducing query response times from days to minutes.
• The system demonstrates impressive performance, handling complex, multi-join queries on massive datasets in under a second.
• Fetch uses a nightly full data refresh into versioned, wide tables, simplifying LLM integration and data rollback.
• Key benefits include democratized data access, faster iteration for sales/commercialization teams, and freeing up the analytics team for foundational challenges.

---

📖 Source: How Fetch built its flexible agentic analytics tool on ClickHouse Cloud