The misconception
Most teams think of data modeling as a technical step.
Something done after collection.
Something handled by engineers.
Something optional.
The assumption:
If the data is in BigQuery, it’s ready to use.
That’s not what actually happens.
What data modeling actually is
Data modeling is the system layer that defines structure.
Not just moving data.
Defining how it behaves.
This includes:
- shaping tables into consistent formats
- defining relationships between entities
- standardizing metrics and dimensions
- enforcing reusable logic
This is where data becomes predictable.
Structure defines system behavior.
Modeling is where that structure is created.
Why raw data isn’t usable
Raw data reflects how systems collect information—not how it should be interpreted.
In systems like Google Analytics 4 exports:
- data is nested
- events are inconsistent
- logic is implicit
- relationships are undefined
This makes it difficult to:
- write reliable queries
- define consistent metrics
- compare results over time
If you query raw data directly, you’re rebuilding logic every time.
Without defined context, every query becomes an interpretation.
And interpretation is where inconsistency begins.
This is where systems begin to drift.
For a deeper breakdown, see How GA4 BigQuery Export Changes Everything.
What modeling actually does
Modeling introduces structure into the system.
Not as a step—but as a foundation.
1. Defines entities
Instead of raw events, you work with:
- users
- sessions
- transactions
These become stable units of analysis.
2. Standardizes logic
Metrics are defined once—not recreated in every query.
This ensures:
- consistency across reports
- alignment across teams
- repeatable outputs
For where this logic should live, see Where Logic Belongs in a Data Estate.
3. Creates relationships
Modeling defines how data connects.
Without it:
- joins are inconsistent
- results conflict
With it:
- relationships are predictable
- queries become stable
4. Simplifies querying
Queries no longer reconstruct logic.
They express intent.
This is what allows systems to scale.
AI reduces the effort required to query data.
It does not reduce the responsibility of structuring it.
Modeling and AI
AI does not model your data.
It queries it.
AI does not understand your data.
It relies on how your system defines it.
If your data is not modeled:
- structure must be inferred
- queries vary unpredictably
- outputs become inconsistent
This is why modeling is not optional.
It defines what AI is able to interpret.
For how this appears at the interface, see Conversational analytics.
Modeling vs pipelines
This is where confusion happens.
Pipelines move data.
Modeling defines it.
You can have:
- perfect pipelines
- automated ingestion
And still have unusable data.
Because:
movement is not structure
For a deeper breakdown, see Data Pipelines vs Data Systems.
Where modeling fails
Modeling doesn’t break.
It drifts.
1. Inconsistent definitions
Metrics change over time.
Different queries produce different answers.
2. Logic duplication
The same logic exists in multiple places.
Each version behaves differently.
3. Fragile queries
Small changes break outputs.
Because structure isn’t enforced.
4. Misaligned reporting
Dashboards disagree.
Not because tools are wrong—but because structure is missing.
If this is happening:
the issue isn’t your reports
it’s your model
For how this propagates, see Why AI Analytics Fails.
Where this fits in your system
Data modeling sits between:
- data collection
- and data interpretation
It is part of the processing layer of your data estate.
It does not:
- collect data
- define meaning
It provides the structure that makes both usable.
It connects directly to:
- semantic layer (meaning)
- data agents (execution)
- conversational analytics (interface)
What this enables (when it works)
Modeling doesn’t add capability.
It enables reliability.
- consistent metrics
- reusable logic
- stable querying
- trustworthy outputs
This is what allows:
- reporting to align
- AI to function
- decisions to reflect reality
The value of AI in analytics is not better answers.
It is faster access to answers—if the system is correct.
Connection to AI-ready data
AI-ready data is not possible without modeling.
Because AI depends on:
- predictable schemas
- consistent logic
- defined structure
If modeling is missing:
AI is forced to interpret raw data
And that interpretation is inconsistent.
What to do next
If the same question produces different answers, the issue isn’t the query.
It’s the structure.
Modeling is what makes AI usable
See AI-ready data
Evaluate your system
See Evaluate
Final principle
Raw data reflects activity.
Modeled data reflects structure.
And without structure:
the same question will never return the same answer twice.
Doug McCaffrey
Designs and maintains analytics systems that remain reliable over time.
Explore how this connects across your data estate: