How to Determine the Minimum Granularity of a Table (Tableau Approach)

A visual guide using three perspectives: analytical reasoning, decision flowchart, and mental model.

1. Reasoning Map

Understand what granularity means

To determine granularity you must start with two key questions:

"What does a single row in this dataset represent?"
"Which dimension or combination of dimensions uniquely identifies each record?"

Examples:

  • POS Sales → Store + SKU + Day
  • Inventory → Store + SKU
  • Orders → OrderID + LineID
Uniqueness test in Tableau

Procedure:

  • Drag candidate dimensions to Rows
  • Add CNT(Number of Records) to Text
Filter → CNT(Number of Records) > 1

If no rows appear, the combination uniquely identifies each record.

Remove redundant dimensions

Remove one dimension at a time.

If uniqueness remains unchanged, that dimension is redundant.

Store + SKU + Date + Country
Store + SKU + Date

Conclusion: Country is derived from Store.

Identify dependencies between dimensions

Use COUNTD to check whether one dimension functionally determines another.

COUNTD(Store)
COUNTD(Store, Country)

If the results are equal:

Store → Country

Country is therefore derived information and does not add additional granularity.

2. Decision Flowchart

Start with all candidate dimensions
↓
Place them in Tableau Rows + CNT(Number of Records)
↓
Are there records with CNT > 1 ?
↙
YES
Add another dimension
↺ Click to highlight and loop back to decision: CNT > 1 ?
↘
NO
The combination uniquely identifies records
↓
Remove one dimension
Does uniqueness remain?
↙
YES
Redundant dimension
Remove it
↓
↘
NO
Required dimension
Keep it
↓
All dimensions tested?
↙
YES
Remaining dimensions = Minimum Granularity
↘
NO
Continue testing
↺ Click to highlight and loop back to decision: Does uniqueness remain?

3. Mental Model (Mind Map)

Understand what a row represents
Identify candidate dimensions
Minimum Granularity
Uniqueness test (CNT)
Remove redundant dimensions
Check dependencies (COUNTD)
Independent dimensions

Signals that granularity is correct