Share with your CDO
Enterprises deploying AI on top of unresolved data governance gaps are building on sand, and the analytics engineer role is the missing structural fix. The argument is that data engineers move data, data scientists model it, and data analysts report on it, but nobody owns what the numbers actually mean across systems. That ownership vacuum is why finance, product, and an LLM can each return a different revenue figure for the same period, with no single team accountable for reconciling them.
What this means for your business
The “which number is right” problem isn’t a data quality problem. It’s an ownership problem, and AI makes it expensive. Manual dashboards could absorb some definitional drift because a human analyst might notice the discrepancy before it reached the boardroom. When an LLM is querying the same inconsistent data layer and synthesizing answers at scale, that drift compounds silently. Organizations that have already shipped AI-powered reporting or decision-support tools without resolving metric ownership are the ones this story is about.
The analytics engineer, as described here, owns the semantic layer, meaning the governed, version-controlled definitions of every metric the business runs on. Think of it as the difference between a database knowing that a column is called “revenue” and the business actually agreeing on what counts as revenue across product lines, geographies, and regulatory jurisdictions. The role embeds validation logic directly into pipelines so bad data stalls before it lands, rather than triggering an alert after a corrupted figure has already been consumed. That’s a materially different failure mode than most enterprises currently operate with.
The argument holds structurally, though it’s worth noting that writing for a CIO-focused outlet gives the author an incentive to frame org-chart gaps as solvable through new headcount rather than through better tooling or sharper ownership mandates on existing roles. The semantic layer problem is real. Whether it requires a dedicated hire versus a reassigned senior data engineer with broader authority is the question CDOs should push on before opening a requisition. If your current data team can’t describe who would be fired if a metric definition was wrong, that’s the diagnostic, not the job title.
Concept deep-dive: Semantic layer
A semantic layer sits between raw data and the tools that consume it, translating physical database tables into business-meaningful definitions like “monthly recurring revenue” or “active user.” It exists because the same underlying data can be sliced dozens of ways, and without a single governed definition, every team builds its own interpretation. The business consequence is that AI systems querying that layer inherit whichever definition they happen to hit, making consistency a structural property, not a review process.
Based on reporting from The missing role in every enterprise AI strategy: The analytics engineer, originally published 2026-08-03 07:04:00.
