№ 03 DEC 04, 2025 MEASUREMENT 7 MIN READ

The post-attribution playbook

iOS 14 broke last-click. GA4 is imperfect. Third-party cookies are functionally dead. The attribution model you relied on for a decade is gone, and nothing has cleanly replaced it. Here's what actually works now — and it's older than you think.

For fifteen years, digital marketing had a cheat code that no other advertising medium ever had: click-level attribution. You could trace a dollar of ad spend to a dollar of revenue with a straight line. TV couldn't do that. Radio couldn't do that. Billboards couldn't do that. Digital could, and it made us lazy.

We stopped asking whether our marketing worked. We only asked whether our tracking said it worked. Those are not the same question. The gap between the two has been widening since iOS 14.5, and most marketers are still pretending it hasn't.

What we actually lost

The specific thing that broke was deterministic, cross-site tracking. Before ATT, Meta could follow a user from an ad impression to a website visit to a purchase and report that chain back to the advertiser. After ATT, roughly 75% of iOS users opted out. Meta's reported conversions dropped 30–40% overnight — not because the conversions stopped happening, but because the tracking could no longer see them.

GA4 made it worse, not better. The shift from session-based to event-based analytics, combined with shorter attribution windows and aggressive data thresholding, means GA4 routinely underreports paid channel performance by 20–50% in our audits. The dashboard says one thing. The bank account says another.

The result is a measurement vacuum. Most brands we audit are making budget decisions based on data that's directionally right but quantitatively unreliable. They're flying a plane with a compass but no altimeter.

The old methods, revisited

The irony is that the tools we need already existed — they just fell out of fashion because click-level attribution was easier. The post-attribution playbook has three components, all of which predate digital marketing.

First: geo-matched market tests. Pick two statistically similar markets — say, Denver and Portland. Run ads in one, dark the other. Compare revenue. This is the cleanest incrementality signal available because it doesn't rely on any tracking pixel. It uses the business's own revenue data as the source of truth. We run these quarterly for every client spending above $50k/month.

Second: holdout groups. Suppress 10–15% of your retargeting audience and compare their conversion rate to the exposed group. This tells you the true incremental value of retargeting — which, in our data, is typically 40–60% lower than what the platform reports. Most brands are paying for conversions that would have happened anyway. Holdout groups quantify how much.

Third: media mix modeling. MMM uses regression analysis to estimate the contribution of each channel based on spend and outcome patterns over time. It doesn't require user-level tracking. The catch is it needs 18–24 months of clean historical data and enough spend variation to detect signal. For brands spending $100k+/month, it's the best strategic tool available. For brands spending less, it's noise.

Attribution didn't die. It just stopped being automatic. The brands that measure well in 2026 are the ones willing to do the work that click-tracking used to do for free.

The practical stack

Here's what a working measurement stack looks like in practice for a brand spending $50–200k/month across three or more channels:

No single layer is sufficient. The stack works because each layer compensates for the weaknesses of the others. Platform data is fast but inaccurate. Geo tests are accurate but slow. Server-side tracking is a good middle ground. MMM is strategic but lagging.

The organizational challenge

The hardest part isn't technical. It's cultural. Most marketing teams have been trained to expect real-time, click-level attribution. Telling a CMO that the true impact of a campaign won't be measurable for six weeks — because that's how long the geo test takes — requires a different kind of trust than showing them a dashboard.

Build that trust incrementally. Start with server-side tracking to close the obvious gaps. Run one geo test to demonstrate the method. Share the results with finance, not just marketing. Once the CFO sees a measurement method they trust more than platform dashboards, the organizational buy-in follows.

The attribution era was convenient. The post-attribution era is more honest. We're measuring what actually happened, not what a pixel said happened. That's harder. It's also better.

← All essays Next: Why we fired our biggest client →
→ ATTRIBUTION BROKEN?

Start with
the audit.

We'll map the gap between what your dashboards say and what your bank account says. The memo is free. The fix usually isn't — but it pays for itself.

Reply to the author → Or read the next one →