Data Modelling
Part of Business intelligence architecture for growing companies
Deciding which reports need a shared data model
Use meaning, reuse, access and change impact to decide which reports should share a data model.
Reports are candidates for a shared data model if they repeatedly use the same business entities or measures and their figures must agree. A short investigation for one audience may be easier to keep local. Decide by weighing meaning, reuse, access and the impact of changes on dependent reports.
Check the overlap
List each report's audience, decision, source, data grain, measures, filters and owner. Compare the calculations, not just their names: two figures labelled “sales” may count different events or use different periods. Resolve those differences before giving them one shared name.
A weekly operations report and a monthly management report might both use an approved count of completed cases, grouped differently. That common count could live in a shared model while each report keeps its own presentation.
Key considerations for shared data models in Australia
- Audience overlap
- Multiple teams use similar metrics (e.g., sales, case completions)
- Measure consistency
- Same definitions (e.g., 'approved completed cases') across reports
- Data grain alignment
- Common level of detail (e.g., daily, weekly, by region)
- Ownership clarity
- One owner for shared calculations and relationships
Apply four tests
| Test | Sharing may help when… | A separate model may be clearer when… |
|---|---|---|
| Meaning | Reports need the same approved entity or measure definition. | Similar labels conceal different populations or rules. |
| Reuse | Maintained reports use the same fields and relationships. | The work is exploratory or has one limited audience. |
| Access | The governed data can safely serve the intended users. | One audience needs detail others should not receive. |
| Change | An owner can assess dependent reports before changing a measure. | Changes are frequent and specific to one experiment. |
No reliable report-count threshold exists. Nor should every subject be forced into one model. Microsoft's managed self-service Power BI guidance describes a blended approach: a centralised team maintains shared semantic models, while report creators in business units reuse them. This is product guidance, not a rule for every platform.
When to share a data model vs. keep it local
- Meaning
- Reports need the same approved entity or measure definition.
- Reuse
- Maintained reports use the same fields and relationships.
- Access
- The governed data can safely serve the intended users.
- Change
- An owner can assess dependent reports before changing a measure.
Pros and cons of shared data models in Australian organisations
- ProsConsistent reporting across departments; supports ATO and GST compliance via uniform measures; reduces duplication in superannuation and payroll reporting
- ConsOverly rigid models may hinder agile analysis; sensitive data exposure risk if access controls fail; change impact harder to track in large teams
Define what the model covers
State the business questions the model supports, the grain of its key tables, approved measures and exclusions. Specify the date each measure uses: grouping orders by placed date can give a different weekly series from grouping them by completed date.
Give shared calculations and relationships an owner. Report owners need a way to request fields and learn about changes that affect meaning.
A dependency list shows which reports need review; their results still need checking. The shared model acts as a data source for report creators, so its scope affects what authors can build.
Account for access
A model may expose more than any one report displays. Protect sensitive fields with appropriate access controls or exclude them from the model. Check permissions in the platform and configuration actually used.
Record the choice
For each report, record whether it will use an existing shared model, help justify a new or extended model, or retain a local model. Note the reason, owner and event that would prompt reconsideration. Before moving a report, compare selected results for the same period and population, then review intentional differences with its owner.
Steps to decide on a shared data model
- Record whether the report uses a shared model, new/extended model, or local model
- Note the reason, owner, and event that would prompt reconsideration
- Compare selected results for the same period and population before migration
- Review intentional differences with the report owner
