The iOS 17 Attribution Playbook: What Still Works in 2026

Five years after ATT shipped, most brands are still lying to themselves about where their revenue comes from. Here's the measurement stack we run at every client above $250K/mo.

iOS 14 was a warning. iOS 17 was a declaration. iOS 18 killed the debate.

If you are still using last-click, platform-reported ROAS in 2026 to make budget decisions, you are flying blind. Here is the stack we build inside every engagement above $250K/month.

Layer 1 — Server-side event foundation

Pixel-only is over. You need:

Meta CAPI via a gateway (Stape, Elevar, or a custom Node endpoint). Event Match Quality above 8.0.
Google Enhanced Conversions through GA4 server-side.
TikTok Events API — same story.

Dedupe everything. A client who is double-counting will scale into the wall.

Layer 2 — Blended attribution, not platform-blessed

We deploy one of Northbeam, TripleWhale, or (for very large accounts) a Rockerbox + custom BI stack.

The job: a blended view that reconciles what the platforms claim vs what the P&L records. The truth is somewhere in between, and the gap is where your budget decisions live.

Layer 3 — Incrementality testing on a rolling schedule

Every quarter we run:

Geo-holdout tests on at least one major channel. If Meta is 55% of spend, we turn it off in three test markets for two weeks.
Conversion-lift studies when the platform natively supports them.
Budget-pacing experiments — the same spend, moved across channels in an ABBA design, to measure marginal lift.

This is the only way to know whether a channel is causing revenue or catching it.

Layer 4 — MMM for the $10M+ brands

Once you are spending $500K+/month across three or more channels, platform attribution is too noisy. We build a marketing mix model — typically a Bayesian MMM via Meridian or Robyn — that gives you:

A steady-state understanding of each channel's contribution.
Saturation curves that tell you when the next dollar stops paying back.
Budget allocation scenarios that are defensible to a board.

The trap most founders fall into

Buying a dashboard is not the same as having measurement.

A Northbeam subscription with bad CAPI and no incrementality testing is still fiction — prettier fiction. The work is the setup and the rolling validation, not the screen you look at.

Spend the first month on plumbing. Spend every subsequent quarter on testing. That's the job.