Semantic Web Academy — home

The 90-Day Rollout Playbook

A pragmatic week-by-week plan to ship a real semantic layer without boiling the ocean.

0/2 done

Overview

Don't model the company; model one fact table

Every failed semantic-layer rollout starts the same way: a task-force tries to model every domain at once. After three months they ship nothing. The successful rollout is deliberately small.

The 90-day plan

Weeks 1-2 — Pick the keystone fact. Choose one fact table whose metrics drive the most disputed questions (usually orders, events, transactions or sessions). Ship one semantic model and 5 metrics with owner: and description:.

Weeks 3-4 — Replace one dashboard. Re-point ONE existing executive dashboard at the semantic layer. Verify numbers match (or that the new ones are correct and the old were wrong — that conversation is the prize). Add the dashboard's URL to the metric's metadata.

Weeks 5-6 — Add the second consumer. A notebook, a Streamlit, an embedded chart — anything that isn't the same BI tool. This is when the layer stops being 'a Tableau backend' and becomes a real semantic interface.

Weeks 7-8 — Contracts & ownership. Add metric_contract.yaml for every shipped metric. Wire CI to fail PRs that change a definition without bumping the version.

Weeks 9-10 — Pre-aggregations. Instrument query traffic, find the top 5 query shapes, build pre-aggregations for them. Publish the rollup-hit-rate dashboard.

Weeks 11-12 — Activate. One reverse-ETL sync, or one LLM-tool integration. Prove the metric reaches a non-BI surface unchanged.

Week 13 — Retro & roadmap. What 5 metrics ship next quarter? Which fact table after orders? Who is the next owner?

What to refuse

  • 'Let's model everything in OWL first' — slow path to nothing shipped. Start tabular; bridge to ontology later if scale demands it.
  • 'Let's switch BI tools at the same time' — two tool migrations at once is one tool migration that fails twice.
  • 'Just use views in the warehouse' — that's where the five-revenues problem starts. The semantic layer must be a compiler, not a shelf of SQL.

Analogy — pave the cow paths

A 90-day rollout is paving the cow paths, not master-planning the whole city. Ambitious teams try to design every road before pouring a metre of concrete, and a year later still have only a beautiful map. The teams that ship watch where people already walk — the one fact table behind the most-argued-about number in the company — and pave that path first. One keystone fact, five metrics, one migrated dashboard: a short stretch of real, drivable road beats a gorgeous plan of roads nobody can use yet. You earn the right to pave the next path by finishing this one.

Reflect

Every previous lesson in this track has been background for one question: what is the smallest, most legible version of a semantic layer you could ship in 90 days? Pin it to one fact table, five metrics, one dashboard, one second consumer. Anything bigger than that on day one is over-engineering. Anything smaller is theatre.

  • ›What's the keystone fact table you would pick today, with everything you've learned?
  • ›Who are the five owners you would assign — and have you spoken to them yet?
  • ›What's the ONE dashboard whose migration would prove the model to skeptical executives?

Reading in progress · 0 of 2 activities done