How to Assess Data Readiness for AI: Snowflake's 6C Framework

Most AI projects fail due to data, not models. Unlike traditional analytics, where poor data leads merely to incorrect charts, in AI the same defects become embedded in indexes, are retrieved, and produce confident but false answers, potentially leading to erroneous actions. As system autonomy increases, the consequences of errors amplify. For systematic assessment of data readiness, the "6C" framework is proposed: Clean, Contextual, Consumable, Current, Correlated, and Compliant. Each of the six factors is detailed for different workload types: Retrieval-Augmented Generation (RAG), agents, feature services, and model training. Snowflake implements the 6C requirements through its tools: Data Metric Functions for quality control, Semantic Views for context, Cortex Search for vector search, Feature Store for storing features, Streams and Tasks for currency, Time Travel and ACCESS_HISTORY for correlation, and Horizon Catalog for compliance. The framework is packaged as an open-source agent skill that runs data assessments using over 60 metrics for a given workload scenario (e.g., service or training). The agent has read-only access, and any data modification requires explicit approval. An example evaluation showed: the Clean factor passed, while Contextual failed (with low scores for semantic documentation). The project repository is on GitHub: github.com/Snowflake-Labs/ai-ready-data.