← Blog / ATTRIBUTION
Attribution in 2026: what iOS broke stayed broken
18 June 2026 · 2 min read · by Obscale
LOG / attribution-2026
operating notes
In April 2021 Apple shipped App Tracking Transparency and the performance marketing industry split into two camps: teams that rebuilt their measurement, and teams that learned to live with dashboards that lie politely.
Five years later the split is permanent. Here is the state of attribution in 2026, as we see it from inside our own e-commerce operation.
The platforms adapted by modeling
Meta, Google, and TikTok all responded to signal loss the same way: fill the gaps with modeled conversions. The models are good. They are also graded by the same platform that spends your budget, which is a conflict of interest so normalized that nobody mentions it anymore.
Platform-reported ROAS in 2026 is best understood as a directional instrument, useful for optimization inside one channel, useless for allocation across channels. Every platform claims credit generously, and if you add up their numbers you routinely get more revenue than your bank account does.
Last click died quietly
The industry spent a decade arguing about last-click attribution. Signal loss settled the argument by making last click physically incomplete: the click IDs get stripped, the cookies expire in days, the journey crosses devices. What remains of last click is a report that punishes every channel that works early in the funnel.
What actually works now
The teams doing this well in 2026 converged on the same architecture, whatever tools they use:
First-party data capture at the edge, server-side, with your own pixel on your own domain. Consent-aware by design, because EU enforcement stopped being theoretical years ago.
An identity graph that stitches anonymous sessions, emails, phone numbers, and orders into one profile per human. This is the hard part, and it is the part we spent three years building for ourselves. Ours resolves 4B+ events a month at sub-50ms p99, and it exists because nothing we could buy in 2021 survived our volume.
Triangulation instead of one source of truth: attribution for speed, incrementality tests for truth, MMM for the board. Any one of the three alone will mislead you within a quarter.
The uncomfortable summary
Attribution in 2026 is not a solved problem, it is a managed one. The signal never came back. The teams that win treat measurement as infrastructure they own, not a report they download.
That is the reason ScaleTrack exists. We needed it before we sold it.