For about a decade, marketing attribution ran on a trick: third-party cookies let ad platforms follow one browser across the whole web, so the industry could pretend to watch complete customer journeys, ad view to purchase, across sites and sessions. The trick is over in most places that matter: Safari and Firefox block third-party cookies outright, Apple locked down the mobile equivalent, and while Chrome famously walked back its removal plans, building strategy on the one browser that still permits yesterday is not a strategy; it is a countdown with extra steps.
So what does attribution look like when you stop waiting for the old world back? Four layers, in order of how much you should trust them.
Layer 1: first-party measurement, done with discipline
Everything observable on your own properties survived untouched: that is the whole point of first-party data. Which channel a session arrived from, what it did, whether it converted. The decay is only in cross-site stitching, which means the humble UTM parameter quietly became the backbone of attribution again, and most companies treat theirs like a junk drawer.
Discipline here is cheap and compounds: one written UTM convention (channels, sources, naming, enforced like the event naming convention, because it is one), campaign links generated from a shared tool rather than freehand, and a verified funnel so what arrives gets recorded. This layer answers "which channels bring sessions that convert on my site" with real evidence. It cannot see view-through effects or cross-device hops. Know both facts and it serves you honestly.
Layer 2: server-fed conversion APIs
The platforms lost visibility too, so they built pipes for you to hand them your conversions directly: Meta's CAPI, Google's enhanced conversions, the equivalents everywhere. Hashed first-party identifiers let platforms match your sales to their users server-to-server, restoring their bidding signal and their reporting confidence.
Be clear-eyed about what this layer is: better inputs to interested parties. Feeding platforms fuller data genuinely improves optimization on real budgets, and their resulting attribution claims remain self-graded homework, now with fewer excuses. Run the pipes for the bidding benefit; keep reading the outputs with the standard discount.
Layer 3: media mix modeling
MMM is the pre-cookie grandparent, back in fashion because it never needed user tracking at all: statistical models over aggregate history (spend by channel, seasonality, promotions, outcomes) estimating each input's contribution. No cookies, no consent exposure, no platform dependence, and no user-level anything.
Its honest profile: works at meaningful spend across multiple channels with enough history to model, answers budget-allocation questions ("shift 20% from X to Y?") rather than creative ones, and its outputs are estimates with error bars, not counts. Once an enterprise-only luxury, open-source tooling has pulled MMM within reach of mid-size businesses, and for the "where should next quarter's budget go" question it is now often the most honest instrument available.
Layer 4: incrementality, the only causation on the menu
Everything above measures adjacency. Experiments measure cause: pause the channel in one geography, holdout an audience slice, run the campaign for half, and watch actual orders diverge or fail to. No model, no window, no attribution formula, just the counterfactual, observed. It is slow, it costs test budget, it occasionally executes a beloved channel in front of the team, and one honest quarterly test recalibrates all three layers above it better than any dashboard subscription.
The stack, assembled
The businesses measuring well post-cookie all converge on the same shape: disciplined first-party funnel as the daily instrument, conversion APIs feeding the platforms they pay, MMM (at sufficient scale) for allocation, and a standing habit of incrementality tests as the calibration layer. Notice what is absent: a magical replacement tracker. Vendors selling "the post-cookie attribution solution" as a tag you install are selling the old certainty back to you in new packaging, and the certainty was always partly theater.
The uncomfortable but liberating summary: attribution went from pretending to know each customer's journey to actually knowing your channels' aggregate effects. Less flattering, more true, and entirely buildable from where you stand: the first layer costs a naming convention and an afternoon of verification, which is, not coincidentally, where I start every measurement engagement. Start there this month; add layers as spend justifies them. The companies that made this peace two years ago are not talking about cookies at all anymore. They are talking about what to do with numbers they finally trust.