For most of my career, marketing measurement rested on one comfortable assumption: somewhere, someone could tie a euro of spend to a specific user doing a specific thing. That assumption is gone. Not weakened — gone. And a lot of teams are still building dashboards as if it were coming back.
Every system now tells a different story
The uncomfortable truth of measurement in 2026 is that there is no single number to trust. Apple's SKAdNetwork hands you delayed, aggregated postbacks with a coarse conversion value. Ad networks report their own aggregated, modeled results — flattering, naturally. MMPs no longer pass device-level data the way they used to. Your frontend sees only what the user consented to; your backend sees the transactions but not the marketing touch. Line them up side by side and they disagree. They are all "right" in their own frame.
The tools didn't converge — they fragmented
For a while the industry hoped for a clean replacement. It didn't arrive. In October 2025 Google retired most of the Privacy Sandbox APIs and gave up on a single Chrome-led successor to the third-party cookie. Cookies stay — but consent-gated and unreliable. There is no unified, privacy-preserving standard to migrate to. Instead we got a patchwork: legacy identifiers where they still work, modeled and aggregated signals, and server-side collection for the events you actually control.
Aggregated data is not just "less" data
It's a different shape of data. You lose the join key that let you follow one user across systems, so you can no longer stitch a story at the individual level — only at the cohort, campaign or channel level. That breaks a lot of habits: no more clean last-touch attribution, no more waiting 180 days for a per-user CLV to settle before judging a decision. The question shifts from "which user did this" to "which direction is this pushing the numbers."
This is why marketing mix modeling came back
MMM isn't new — it's what marketers used before deterministic tracking spoiled us. It models outcomes top-down from spend, seasonality and external factors, and it doesn't care about device-level identifiers. Pair it with incrementality testing — geo holdouts, on/off experiments — and you get something identifiers can't give you anymore: a causal read on whether spend actually moved the business. Aggregated inputs, causal conclusions.
Deciding in an imperfect world
So what do you do on a Tuesday, with a budget to allocate? Secure the signals you own: consented first-party events, routed server-side, defined once and trusted. Triangulate — MMP, SKAN, network reports and MMM rarely agree, but where three of four point the same way, that's your signal. And make peace with precision loss. Anyone still waiting for a clean, deterministic number to arrive is waiting for a train that left in 2021. The job was never to measure perfectly; it was to decide well. That part didn't change — it just got honest about the uncertainty it always had.