DATA PRODUCTS · CASE 02

Semantic Model Wiki

An offline-first workflow for documenting metric definitions, DAX, lineage, ownership, and human approval.

My role
Product Owner, governance designer, workflow author
Portfolio status
Sanitized working system

01 / Business problem

The decision behind the work

Semantic knowledge was embedded in models, reports, tribal memory, and conversations. Teams could see a metric name without reliably knowing its definition, formula, source, owner, steward, or whether anyone had approved it.

I designed the documentation so teams could extract it, review it, preserve human edits, and reuse approved definitions across models.

02 / My product decisions

How the decision works

  1. 01
    Preserve definitions across regeneration. I used round-trip markers so human edits survive subsequent metadata refreshes.
  2. 02
    Make ratification a real stage. AI can draft definitions; a person accepts, edits, or defers them before publication.
  3. 03
    Record accountability with the metric. Owner, steward, domain, and sub-domain sit beside the formula and lineage.
  4. 04
    Create a cross-model glossary. Approved metrics are harvested into an offline, syncable asset rather than trapped inside individual reports.

Interactive evidence

Trace a metric from definition to owner

Supplier Scorecard · scrubbed semantic model

Draft definitionDomain: Procurement

Avg Actual Lead Time Days

Average elapsed lead time recorded across purchase-order lines in the selected context.

DAX
AVERAGE(PO[Actual Lead Time Days])
Source
PO · Actual Lead Time Days
Owner
Sourcing & Procurement
Steward
BI Product Owner
Lineage tag
supplier-scorecard / PO / actual-lead-time

A human reviewer must accept, edit, or defer the drafted definition.

03 / Evidence

What the demonstration proves

The scrubbed Supplier Scorecard demonstrates the actual artifact shape: measures, tables, relationships, calculated columns, storage modes, and ownership details in a single inspectable model page.

04 / Outcome

80 / 100+

Roughly 80 documented semantic models yielded more than 100 metrics ratified through stakeholder review.

05 / What I’d carry forward

Extracting metadata makes it visible, not correct. A definition becomes useful only when its owner and approval survive the next refresh.