Most attribution discussions focus on models and tools.
This page explains what actually determines attribution—and why it breaks over time.
What attribution is (and isn’t)
Attribution is meant to answer a simple question:
What contributed to a result?
In practice, it tries to assign credit across:
- channels
- campaigns
- touchpoints
But attribution is not a standalone feature.
It is the result of how your data is:
- collected
- structured
- processed
- interpreted
If those are inconsistent, attribution will be too.
Why attribution breaks
Attribution issues rarely come from a single mistake.
They emerge when systems fall out of alignment.
You start to see it in different ways:
- conversions don’t match across platforms
- revenue doesn’t align with GA4
- channel performance shifts unexpectedly
- reports tell different stories
These aren’t reporting problems.
They’re system problems.
Why platforms don’t agree
Different platforms measure different parts of the same interaction.
- Google Analytics observes on-site behavior
- ad platforms track ad interactions
- backend systems record transactions
Each uses:
- different identifiers
- different timing
- different assumptions
So even when everything is “working,” the numbers won’t match.
Where attribution is actually determined
Attribution is not created in your reports.
It’s determined upstream—by how your system is designed.
Key factors include:
- how events are defined
- how users are identified
- how sessions are structured
- how logic is applied
If these are inconsistent, attribution becomes unstable.
What most teams get wrong
Most teams try to fix attribution by changing models or tools.
They switch between:
- last-click
- first-click
- data-driven models
But models don’t fix inconsistent inputs.
They redistribute them.
This creates the illusion of improvement—without solving the problem.
Why changing models doesn’t fix attribution
When attribution doesn’t make sense, the default response is to change the model.
- last-click
- first-click
- data-driven
This feels like progress.
It isn’t.
Models don’t fix inconsistent inputs.
They redistribute them.
Changing the model changes the distribution—not the truth.
What good attribution depends on
Reliable attribution doesn’t come from a better model.
It comes from a well-structured system.
That includes:
- consistent event definitions
- stable identity handling
- aligned logic across platforms
- complete and accurate data collection
When these are in place:
- attribution becomes explainable
- differences between platforms become understandable
- decisions become more reliable
Why attribution changes over time
Even well-built systems don’t stay stable.
Attribution shifts as:
- tracking implementations evolve
- privacy constraints increase
- platforms change how they process data
- systems grow more complex
Without ongoing maintenance, attribution drifts.
The constraint most teams underestimate
Attribution is increasingly limited by factors outside your control.
Privacy, consent, and browser behavior affect:
- what data is collected
- how users are identified
- how long data persists
This means attribution is not just a technical problem.
It’s a constrained one.
What good attribution actually looks like
Perfect agreement across platforms is not the goal.
Different systems will always:
- observe different parts of the journey
- use different identifiers
- apply different logic
Good attribution is not about agreement—it’s about consistency.
When attribution is working well:
- results are internally consistent
- definitions are clear
- differences are explainable
- decisions are based on stable signals
Why attribution changes over time
Attribution doesn’t remain stable.
Even if you don’t change anything.
The system behind it is constantly evolving:
- browser and privacy changes reduce signal coverage
- consent behavior changes what data is available
- platforms update how data is processed
- tracking implementations drift over time
These changes are often gradual and invisible.
Attribution doesn’t fail all at once—it drifts.
What this leads to
If attribution doesn’t make sense, the issue isn’t the model.
It’s the system behind it.
Reliable attribution doesn’t come from models or tools.
- structured data collection
- consistent logic
- aligned systems
- ongoing maintenance
Not from tools alone.
Where tools fit
Analytics tools don’t determine attribution.
They reflect the system they operate within.
To understand how platforms like Google Analytics, BigQuery, and Tag Manager actually fit together:
The next step
Before trying to improve attribution, you need to understand how your system is actually behaving.
An Evaluate engagement provides a structured view of:
- where attribution is breaking
- how discrepancies are introduced
- what is required to restore consistency
Start with Evaluate.
Doug McCaffrey
Designs and maintains analytics systems that remain reliable over time.