Part 5 · Tools of the trade · 4 of 6
Seeing the Data: Tableau, Power BI, Looker, Qlik, Metabase, Redash
BI tools are the consumption layer. The important difference between them is where the metric definition lives: in each dashboard, or in a governed semantic layer everyone reads.
3 min read
Remember week 2: a dashboard is not the data product; it reads one. The question to ask of any BI tool is where does the definition of revenue live? Tools split into two camps.
| Tool | Camp | How you use it | Strengths / cautions |
|---|---|---|---|
| Tableau (Salesforce) | Visual exploration first. Definitions can live in Tableau data sources but often live in each workbook. | Connect to a table, drag dimensions and measures onto shelves, build sheets, assemble a dashboard, publish to Tableau Cloud/Server. | Best-in-class visual analysis; risk of metric drift across workbooks; licence per user. |
| Microsoft Power BI | Visual first, with a modelling layer (DAX measures, Power Query). Now part of Fabric. | Import or DirectQuery a model, write DAX measures, build report pages, publish to a workspace. | Ubiquitous in Microsoft shops; cheap per user; DAX has a learning curve; models can sprawl. |
| Looker (Google) | Semantic layer first. LookML defines dimensions and measures in version-controlled code; every explore and dashboard reads them. | Model tables in LookML (views, explores, measures), then users explore without writing SQL. | Governed metrics by design: the tool closest to "one definition, many consumers". Heavier setup; developer skills needed. |
| Qlik Sense | Associative engine: in-memory model where every selection filters everything. | Load data into a Qlik app, build sheets, use associative filtering. | Strong ad-hoc exploration; definitions live in the app's load script. |
| Metabase | Lightweight open-source BI; questions written by clicking or in SQL; simple dashboards. | Point at the warehouse, ask a question, save it to a dashboard. | Great for small teams; limited governance. |
| Redash | Open-source SQL-first query and dashboard tool (the one in your PRD). | Write SQL, save the query, add visualisations, pin to a dashboard, schedule refresh. | Perfect when everyone can write SQL; definitions live in the query text. |
Semantic layers, in one paragraph
A semantic layer is the place where revenue is defined once as a measure over modelled tables, with its dimensions, and every dashboard, notebook, or AI agent asks it rather than re-computing. LookML is the classic example; dbt's metrics layer, Cube, AtScale, and the semantic model in Fabric play the same role. In the Playbook's terms it's the Metric Dependency Tree made executable. Without one, week 6's "which revenue number is right?" happens every quarter.
Questions a Data PM asks
- Where is this number defined, and can I read the definition? If the answer is "in the workbook," you have a consumption layer pretending to be a product.
- Who owns the dashboard, and who owns the metric under it? Different people, usually.
- How is adoption measured? Distinct viewers per week by role (week 4), not views.
- What happens when the definition changes? One edit in a semantic layer, or twenty workbooks?
Source: Synthesized for DPM Lab from vendor documentation and public positioning (as of 2026).
Where this shows up
1 flashcard in Review