In one line
When you need channel-level contribution without user-level tracking, the method is MMM (Marketing Mix Modeling) — modelling how channel spend, external factors, and baseline demand related to outcomes over time.
It does not track users
Attribution and an MMP follow individual user touchpoints. MMM does not. It reads aggregate weekly data and estimates statistically how much outcomes moved when a channel's spend moved.
That is why it keeps working when user-level tracking is restricted. In exchange, it cannot say who any individual conversion belongs to.
Two things it must account for
- Adstock — today's advertising carrying over across following days
- Diminishing returns — each additional unit of spend producing less
Leave these out and you get an unrealistic model where doubling spend doubles outcomes.
There are clear conditions where it fails
Short data does not work. You need at least a year of weekly observations, preferably two, to separate seasonality.
Channels that move together do not work. Two channels always raised and cut in step cannot have their contributions separated. That is multicollinearity, and per-channel numbers produced in that state cannot justify a budget split. Check with a VIF diagnosis before modelling.
It is not causal
The limitation that matters most. MMM models observed association, not an experiment. A large coefficient does not establish that the channel produced the outcome.
Use it to form hypotheses, and confirm with a holdout test before large budget decisions.
Go deeper
MMM's assumptions and interpretation limits are covered in marketing mix modeling; collinearity diagnosis in the multicollinearity guide. To run it on your own weekly panel, use marketing response analysis.