Attribution

Why systems disagree

1. Different windows. A seven-day and a thirty-day window credit different campaigns for the same sale.
2. Different identity resolution. Platforms match users differently, and none sees the full path.
3. Different conversion definitions. A platform conversion and a CRM closed deal are not the same event.
4. Overlap. Two platforms can each claim the same conversion, so totals across platforms are always inflated.

The working rules

• One system of record for revenue — normally the CRM. Everything else is directional.
• Always state the window and the model beside the number.
• Never sum conversions across platforms.
• Use platform data for optimisation decisions and CRM revenue for budget decisions; keep the two conversations separate.

Choosing a model you can defend

Last-touch is simple and undervalues everything upstream. First-touch is useful for understanding discovery and useless for allocation. Linear and time-decay spread credit in ways nobody can verify. Data-driven models improve with volume and remain opaque.

In practice the honest position is to pick one model, keep it stable so period comparisons mean something, and treat large decisions as requiring corroboration — a spend change that shows up in CRM revenue, not only in a platform column.