Model comparison
One journey. Seven different answers.
That is not automatically a bug. Every attribution model makes an assumption. The problem is pretending one assumption is universal.
Same journey, different assumptions
Do not hide model disagreement.
Show the percentage assigned by each model, events that moved, branches that gained or lost credit, assumptions causing the difference, and the source journey.
Paid search
Content
Retargeting
Sales call
Won
Example data · selected model
First touch
Where did the journey begin?
- Favors
- Acquisition source
- Main limitation
- Ignores everything after entry
Decision guide
Choose the assumption that fits the decision.
| Use case | Models to consider | Question to validate |
|---|---|---|
| Simple operational rules | First / last touch | Where did the journey begin or end? |
| Directional, explainable allocation | Linear / time decay | What does a transparent weighting show? |
| Transitions or combinations | Markov / Shapley | What changes when a state or combination changes? |
| Hierarchical systems | Hierarchy-preserving | What information is lost when structure is flattened? |
No model gets to call itself truth. It earns usefulness by changing a decision and surviving validation.
Run a model comparison.
Bring one journey, inspect the assumptions, and see where the answers diverge.