Notes from the practice.
Notes on data leadership, methodology, architecture, and what 25+ years of analytics transformation taught us.
Nine posts on the most important infrastructure decision of 2026.
Published on the DirectionalData company page on LinkedIn over nine weeks. Each card below links to the live post.
The semantic layer just became infrastructure.
Five years ago this was a niche analytics-engineering term. As of January 2026, Gartner is calling it non-negotiable infrastructure on the same line as data platforms and cybersecurity.
Eight supporting posts.
The four-tools problem.
Forrester 2026: 61% of enterprises run 4+ BI tools and 25% run 10+. 84% of teams still hit conflicting versions of the same metric. The fix isn’t fewer tools — it’s one definition layer they all read from.
Read on LinkedIn →AI accuracy is a metric definition problem.
Text-to-SQL drops 50%+ on real enterprise warehouses. It climbs back to 90%+ with a semantic layer between the model and the data. The model isn’t the bottleneck. The metric layer is.
Read on LinkedIn →Day-2 architecture for the semantic layer.
Most semantic layers we see were designed for the company the team had eighteen months ago. Three Day-1 mistakes we see constantly, three questions to ask before naming a metric, and why rebuilds cost 5–10× what foresight does.
Read on LinkedIn →Stabilize, Improve, Leverage — applied to the semantic layer.
The methodology we’ve used for twenty years, mapped onto semantic-layer work. Most failed projects we inherit tried to Improve and Leverage simultaneously, before stabilizing. Here’s why the order matters.
Read on LinkedIn →Why the CFO should own the metric layer.
In most mid-market companies the metric layer sits under the CIO by default. That’s an org-design accident. Metric definitions are control activities — especially the moment an AI assistant starts answering executive questions.
Read on LinkedIn →The first 90 days of a semantic layer program.
Stabilize the conversation (weeks 1–2). Pick the platform and define ten metrics (weeks 2–6). First wave of BI + AI consumers (weeks 7–11). Stewardship live, backlog expanded (weeks 11–13). The pivot point is the first meeting where dashboards don’t disagree.
Read on LinkedIn →Open Semantic Interchange — what mid-market buyers should know.
A vendor-neutral standard finalized in January 2026. Sixteen-plus signatories. Metric definitions are now portable across platforms. The metric-rebuild portion of platform switching cost just got smaller.
Read on LinkedIn →What multi-grain semantic modeling looks like in practice.
One published Tableau Data Source served four very different audiences at an eight-hundred-restaurant chain, from a manager’s tablet to a multi-year strategic view, all from one set of definitions. The design that made it work: deliberately variable grain, hourly by item for the last ninety days and progressively coarser further back. A concrete look at the pattern, and why it ports straight into Snowflake and Databricks in 2026.
Read on LinkedIn →