Understand what changes when you switch from first touch to linear, data-driven, or a custom model.
Keep the outcome fixed
An order does not change because an attribution rule changes. What changes is the distribution of credit across the touchpoints recorded before it. Holding the outcome fixed helps you see what the model actually contributes.
Imagine a $300 order with seven recorded touches. A linear rule divides the credit equally. First touch allocates it to the earliest recorded interaction, while last touch allocates it to the latest. None of those choices creates a new sale.
Read the rule before the result
Time decay puts more weight on recent touches. Position-based methods usually reserve more credit for the first and last touches. A data-driven method uses a fitted model rather than a fixed allocation pattern; its behavior depends on the data and method used.
A custom model makes room for a stated allocation of your own. Its assumptions should be as inspectable as any other method. A model name alone does not explain the result. The inputs, windows, exclusions, and weights matter.
Compare with a purpose
Use a model comparison to understand how sensitive a conclusion is to the credit rule. If a channel looks strong under one rule and weak under another, that difference is a question to investigate.
Treat incomplete journeys carefully. Unobserved interactions can change the apparent first or last touch. Choosing a model is choosing a lens on the recorded evidence, rather than proving which campaign caused the purchase.
What is the source? What is the method? What remains unknown?
Explore the platform
