The missing layer in the modern data stack (www.dimensionlabs.io)

The article argues the Modern Data Stack (MDS) has a blind spot: it’s optimized for structured, tabular data while more than 80% of enterprise data — chat transcripts, call recordings, emails, reviews, and free-text survey answers — is unstructured and therefore invisible to analytics and BI. The authors propose a new, standardized “Unstructured Data ETL” layer that does for language data what dbt did for tables: ingest from diverse sources (Zendesk, Qualtrics, Gong, Genesys, APIs, Kafka), normalize sessions and metadata, remove or mask PII, and output analytics-ready records into central warehouses (Snowflake, BigQuery, Redshift). Without this layer companies face CX blind spots, slower product iteration, operational inefficiency, weaker ML models, missed revenue signals, and dashboards that show “what” but not “why.” Technically, the layer combines preprocessing (sessionization, deduplication, PII handling) with AI-driven enrichment — LLM orchestration, topic/sentiment classification, intent detection, embeddings and root-cause extraction — mapped to a reusable schema (Category, Sentiment, Root Cause, Product Area, etc.). The result: SQL-queryable, joinable tables that plug into BI and ML pipelines, reduce engineering lift versus ad hoc builds, and unlock high-value signals from previously “dark” language data across the enterprise.