AtScale tested seven leading AI models on enterprise data queries. Without its semantic layer, accuracy sat below 40%. With the semantic layer in place, accuracy improved between 1.7 and 2.4 times across all seven models. The finding confirms what data leaders across the DACH region keep telling us: throwing a more capable model at messy data does not fix the underlying interpretability problem.
A semantic layer translates raw database schemas into business concepts. Revenue becomes ‘revenue by product line and quarter’. Customer becomes ‘active customer with at least one transaction in the past twelve months’. Without that translation, an AI model guessing at column names and join logic will answer confidently and incorrectly. The gap between technical accuracy and business accuracy is where analytics projects stall, and where companies burn months of data engineering time before realising the model was never the bottleneck.
For data teams, this research reframes the hiring conversation. The scarcest skill sits in the middle ground: people who can translate raw data into business concepts, define what metrics actually mean, and build the layer that makes AI useful rather than merely impressive. We have placed candidates over the past year whose primary value was exactly this: semantic modelling, data governance, and the patience to document business logic that nobody had written down.
Companies that have not invested in a semantic layer, or in the people who can build and maintain one, will find themselves cycling through analytics tools and AI assistants without improving the numbers that matter. The multiplier effect AtScale measured points to a discipline that companies can adopt with any tool: defining business concepts before asking AI to answer questions about them.
Prompted by reporting from BigDATAwire.