Run a simple test on your own dashboards: add up the conversions that Meta, Google and TikTok each claim for last month, and compare the total with the orders in Shopify. On most accounts we audit, the platforms collectively claim the month's revenue with room to spare. Every channel is marking its own homework, every channel is grading generously, and the budget meeting is arguing over fiction.
Attribution in 2026 is not about finding the one true model. It is about building a measurement stack whose distortions you understand well enough to make budget decisions you can defend. We run seven-figure annual ad spend across Meta, Google, TikTok and Taboola with attribution in GA4 and a dedicated tracking layer, and this is the mental model behind that stack.
Why the platforms disagree, structurally
Each ad platform attributes conversions using its own window, its own matching logic and visibility only into its own touchpoints. A buyer who clicks a TikTok ad, later searches the brand and clicks a Google ad, then converts from a Klaviyo email will be claimed, quite sincerely, by three systems. None of them is lying; each is answering a different question, namely "did this conversion touch me?", while you are asking "what caused this conversion?"
Layer on top of that the reality of the post-cookie, post-consent web: browser tracking prevention truncating cookies, consent banners removing a share of users from measurement entirely, and modelled conversions filling the gaps with platform-estimated numbers. The result is that platform-reported return on ad spend is a directional signal about creative and audience performance inside that platform, and nothing more. Treating it as a cross-channel truth is how budgets migrate to whichever platform models most aggressively.
The three layers of a workable stack
A measurement stack you can run a business on has three layers, each answering a different question.
Layer one: the ledger. Shopify orders and your accounting system. This is ground truth for how much was sold, at what margin, to new versus returning customers. Every other number in the stack must reconcile against it, and any report that cannot be reconciled is decoration.
Layer two: the behavioural record. GA4, fed properly. Its job is not to be perfect but to be consistent: one measurement system observing all channels with the same rules, so channels can be compared with a shared distortion rather than each channel's private flattery. Getting this layer right is mostly hygiene: clean UTM discipline on every paid and owned link, conversion events defined once and mirrored server-side where possible, and consent mode configured deliberately rather than by default.
Layer three: the decision layer. Incrementality checks and a tracking platform where the business model demands it. Geo holdouts, spend pauses on brand search, or switching off a retargeting audience for two weeks answer the question platforms cannot: what would have happened without the spend? These experiments are unglamorous and occasionally uncomfortable, which is exactly why they are informative.
Server-side tracking: worth it, with expectations managed
Server-side event delivery, whether through a tag manager server container or a conversions API integration, restores a meaningful share of the signal that browsers now block, improves match quality for the platforms' optimisation, and gives you control over what is shared. For a store spending seriously on paid acquisition it is usually worth implementing.
Two expectations need managing. First, server-side tracking is signal repair, not attribution truth: the platforms still grade their own homework, they just see more of it. Second, it does not remove consent obligations; what you send server-side is still governed by what the user agreed to. Implement it as an engineering project with a data map, not as a checkbox in an app.
Running budget decisions from an imperfect stack
The practical operating rules we hold:
- Compare channels only inside one measurement system. Rank channels within GA4, or within your tracking platform, never by mixing Meta's number for Meta with Google's number for Google.
- Reconcile monthly against the ledger. Total attributed revenue across your reporting should be explainable against Shopify orders. Where it is not, understand why before trusting any channel-level story.
- Use platform metrics for what they are good at. Creative comparison, audience diagnostics and delivery health inside the platform. Not cross-channel allocation.
- Schedule incrementality tests. One structured experiment per quarter on your biggest line item of spend will teach you more than a year of dashboard staring.
- Write down the attribution decisions. Which windows, which model, which system settles disputes. Attribution arguments recur forever unless the rules are documented and boring.
The uncomfortable conclusion
There is no configuration that makes attribution precise again. The web that allowed deterministic user-level tracking is gone, and it is not returning. The competitive edge now belongs to teams that accept measured uncertainty and build a decision process around it, rather than teams that keep shopping for a dashboard that promises certainty. An attribution stack is not a product you buy. It is a discipline you run.
Key takeaways
- Platform-reported ROAS answers "did this touch me?", not "what caused this?". Never sum platform numbers.
- Build three layers: the ledger as truth, one consistent behavioural system across channels, and incrementality tests for the big decisions.
- Server-side tracking repairs signal and is usually worth it at scale, but it does not create attribution truth and does not bypass consent.
- Document the rules, reconcile monthly, and let one system settle disputes.