The number on a UA dashboard used to feel like an answer. You looked at CPI, ROAS, and attributed installs, picked the row, allocated the budget. In 2026, that’s not enough. Apple’s SKAN framework still delivers postbacks, but with privacy thresholds and modelled signals. Self-attributing networks each claim credit on their own terms.
MMP dashboards reconcile what they can, but reconciliation isn’t causation. The result is that most app marketers are making seven- and eight-figure budget decisions on numbers that report what users did, not what their campaigns caused.
This isn’t a measurement problem you can patch with another integration. It’s a structural shift in what “performance” can be observed. If you’re running a diversified channel mix across programmatic DSPs, OEM stores, influencer, contextual, and AI-driven discovery, the question gets even sharper: which spend is moving the needle, and which is taking credit for installs that would have happened anyway?
Incrementality testing is the only discipline built to answer that. And in 2026, it’s quietly become the deciding factor in how the most disciplined UA teams are allocating budget.
Why attribution stopped being enough
Attribution and incrementality answer different questions. Attribution asks: among the users who installed, which touchpoint do we credit? Incrementality asks: would those installs have happened without this spend?
For a decade, the gap between those two questions didn’t really matter. Last-click models were good enough. IDFA-level identity made cross-channel deduplication possible. The cost of being wrong was small. Three things changed that.
The signal layer thinned
Apple’s SKAN, Android’s Privacy Sandbox, and high ATT opt-out rates pushed real-time attribution into modelled territory. MMP dashboards started reporting modelled installs alongside deterministic ones. That’s useful, but a confidence interval is not a causal claim.
…