Marketing mix modeling predates digital advertising by decades and was widely considered outdated during the tracking era. American advertisers have returned to it for a specific reason.
The method works on aggregates
The model relates a business outcome, usually sales over time, to the spending and activity levels of each marketing channel across the same periods.
It also incorporates factors outside marketing's control, such as pricing, distribution, seasonality and broad economic conditions, so their effects are not credited to advertising.
Because it uses totals rather than individual records, it never needs to know who saw an ad or link a specific impression to a specific purchase.
User-level tracking became less reliable
Restrictions on cross-site identifiers, platform privacy controls and general fragmentation across devices have reduced how much of a customer journey can be observed directly.
Attribution systems built on following individuals therefore see a shrinking share of reality, and the gaps are not distributed evenly across channels.
A method that never depended on that visibility is unaffected by its decline, which is the core of the technique's renewed appeal.
It measures channels tracking cannot see
Television, audio, print, outdoor and sponsorship generate no click and no user-level record, so click-based attribution assigns them nothing.
Mix modeling treats them the same way it treats digital channels, which allows a genuine comparison between measurable and unmeasurable spending.
This is why the technique is strongest at advertisers who run substantial budgets across both, where attribution systematically favors whichever channel happens to be trackable.
The known weaknesses have not disappeared
The method needs long histories with meaningful variation in spending, and channels that ran at a constant level provide little for the model to learn from.
Correlation between channels that move together makes their individual effects difficult to separate, and the model may distribute credit arbitrarily between them.
It also produces results at a coarse level, answering questions about channel and period rather than about a specific creative or audience.
The two methods answer different questions
Mix modeling estimates how much total effect a channel produced; attribution describes the paths individual converters took before buying.
Treating them as competing versions of the same truth leads to fruitless reconciliation, since they measure different things at different resolutions.
Most large advertisers now run both alongside controlled experiments, using the experiments to check whether either model's estimates hold up when spending is deliberately varied.