Marketing Attribution for Shopify: Why Your Numbers Disagree

Here's a test worth running before you spend another week on this. Add up the revenue each ad platform reports for last month. Compare it to what Shopify says you actually made.

For most stores spending across two or more channels, the platforms claim more than the store earned. Sometimes considerably more.

That result is the whole problem in one number, and it tells you something useful: the reconciliation project you were about to start cannot succeed. You're not looking at one truth measured badly by several tools. You're looking at tools measuring genuinely different things.

Meta counts a sale if someone saw an ad within its window. Google counts the same sale if there was a click in its window. Neither knows the other exists. Both are being honest by their own rules, and both rules are self-serving in the same direction.

So stop trying to make them agree. They never will, and the effort is expensive. What works instead is a hierarchy: one blended number you make budget decisions on, platform data used directionally inside each channel, and real tests when you need to know whether a channel is actually doing anything.

That's a less satisfying answer than a dashboard where everything reconciles. It's the one that survives contact with how measurement actually works now.

Three sources, three different jobs, three different answers. Knowing what each can and can't see stops you asking the wrong one.

Shopify is your source of truth for revenue. Nothing else should be arguing with it about how much money you made. Where Shopify is weak is channel attribution, because by the time an order lands, most of the journey is invisible to it.

Platforms are optimisation tools, not measurement tools. This distinction matters and it's usually collapsed. Meta's data is genuinely valuable for deciding which creative and audience to push inside Meta. It's not a reliable answer to whether Meta deserves more budget than Google — that's a question no platform can answer about itself.

Analytics sits between them and tends toward last-click by default, which systematically over-credits whatever channel closed and under-credits everything that created demand. Branded search is the classic beneficiary: it looks extraordinarily efficient because someone else did the work of making people search your name.

Worth understanding, because it explains why the old playbook stopped working and why nobody has fully replaced it.

Privacy changes reduced deterministic tracking. Mobile operating system tracking restrictions, third-party cookie limits and browser protections mean a large share of journeys can no longer be followed end to end. Platforms responded with modelling — estimating conversions they can't observe. Modelled numbers are not measured numbers, and the gap between them isn't disclosed on your dashboard.

Journeys got longer and messier. Someone sees a TikTok, mentions it to a friend, searches on a laptop three weeks later, then buys from an email. Which channel gets the sale? Every model answers differently and none is wrong exactly.

Dark social is invisible by construction. Recommendations in group chats, private messages, screenshots, podcasts. Real demand creation that leaves no trackable trail at all, and it surfaces as "direct."

Multi-device is normal. Discover on a phone, buy on a desktop. Without a login, those are often two different people as far as your tools are concerned.

The honest consequence: perfect attribution isn't available at any price. Anyone selling you complete visibility is selling modelling with confidence attached. The workable goal isn't precision — it's making good budget decisions despite imprecision, which is a different and achievable target.

Stores burn months arguing about last-click versus first-click versus time-decay. It matters less than the debate implies.

Last click credits the final touch. Simple, and it over-credits closing channels like branded search and retargeting while making top-of-funnel look worthless.

First click credits discovery. Inverts the same bias.

Linear and time-decay spread credit across touchpoints. More reasonable in principle, and still only as good as the touchpoints you can actually observe — which, per the section above, is a shrinking share.

Data-driven models assign credit algorithmically. Better in theory, opaque in practice, and dependent on data volume many stores don't have.

The point everyone misses: every one of these models divides up the touchpoints you managed to track. If a third of your influence is untracked, all your models are carefully allocating credit across an incomplete picture. Changing the model changes who gets credit for the visible part. It doesn't reveal the invisible part.

What to actually do: pick one model, apply it consistently, and use it to spot trends rather than to make precise budget splits. Consistency beats correctness here, because a consistent model shows you movement even if the absolute numbers are wrong — and movement is what you act on.

Four things, in the order most stores should adopt them.

1. Blended metrics as your decision number. Total revenue divided by total ad spend — commonly called MER. It's crude, it can't be gamed by any platform's counting rules, and it answers the question that actually matters: is the whole marketing operation making money? Many operators run everything off blended performance and use platform numbers only inside channels. Watch it against a target rather than chasing per-channel ROAS.

2. Post-purchase surveys. The highest value per unit of effort in this entire article. Ask customers at checkout how they heard about you. The data is messy and self-reported, and it consistently surfaces channels your tracking undercounts — podcasts, word of mouth, dark social. Tools like Kno Commerce and Zigpoll handle this natively. It costs almost nothing and it's the only method that asks the customer directly.

3. Incrementality and holdout testing. The only approach that genuinely answers whether a channel is doing anything. Turn a channel off in a region, or hold out a segment, and see what happens to revenue. Uncomfortable, occasionally expensive, and definitive in a way no dashboard is. Structurally the same discipline as CRO testing, and it needs similar volume to be valid.

4. Marketing mix modelling, at scale. Statistical modelling of how spend across channels drives revenue, including offline and brand effects. It doesn't rely on tracking individuals, which makes it privacy-durable. It needs substantial historical data and real expertise — marketing mix modelling specialists — and it's genuinely valuable past meaningful spend, and premature below it.

Attribution tools consolidate reporting, apply their own modelling, and give you one interface instead of five tabs. Genuinely useful. Just be clear about what they can and can't fix.

What they do well: pull channel data into one place, tie spend to Shopify order data, apply consistent attribution logic across sources, add customer-level and cohort views, and surface blended metrics without a spreadsheet. Options include Triple Whale and Attribuly, with Lifetimely focused on lifetime value and cohorts and Report Pundit for custom reporting.

What they can't do: see journeys that privacy controls made invisible. A tool applying a model to incomplete data produces a confident number from incomplete data — better organised, not more complete. The underlying measurement gap is a property of the environment, not of your reporting stack.

How to evaluate one honestly: ask what its numbers do when they disagree with Shopify, and how it handles conversions it can't observe. A vendor who explains their modelling openly is more trustworthy than one implying perfect visibility.

Worth knowing: attribution tools appear frequently among the apps merchants tell us they're actively evaluating and replacing. In zero-party data from brand operators in the app store research network, 60% reported actively replacing existing apps rather than just exploring, with analytics among the named categories. That churn usually isn't a product failure — it's merchants discovering that a tool couldn't solve a problem no tool can solve, then trying the next one.

Buy one to save time and unify reporting. Don't buy one expecting it to end the disagreement.

Optimising each channel to its own reported ROAS. The most expensive mistake here. Every platform over-reports, so hitting a target ROAS in each channel simultaneously can still mean the business is unprofitable overall. Blended performance is the check.

Cutting top-of-funnel because it shows poor last-click returns. Prospecting creates demand that closes elsewhere. Cut it and you'll watch your efficient branded search quietly dry up over the following months, having removed the thing that fed it.

Treating "direct" as free traffic. Direct is largely untracked demand you paid to create somewhere. It's an attribution gap, not a channel.

Ignoring the profit side entirely. Revenue attribution with no margin view leads to scaling your least profitable products. That requires COGS on landed cost, which most stores don't have — see ecommerce accounting on Shopify.

Judging channels on first purchase alone. A channel with worse acquisition economics can be better on lifetime value. Without cohort measurement you can't see it — covered in retention and lifecycle marketing.

Changing attribution windows or models mid-analysis. Guarantees you can't compare periods, and someone will present the shift as a performance change.

Building dashboards nobody decides from. If a report doesn't change what you do next week, it's decoration.

Your tracking setup is genuinely broken. Distinct from the attribution gap: misfiring pixels, duplicate conversion events, server-side tracking configured badly, or Shopify order data not tying to ad spend at all. This is fixable and worth fixing before concluding attribution is impossible — marketing analytics and attribution experts.

You're scaling spend and decisions are getting expensive. Small budgets tolerate imprecision. Large ones don't, and the cost of allocating badly grows with the number.

You want incrementality testing done properly. Designing a valid holdout — sizing it, choosing regions, avoiding contamination, reading the result honestly — is a real discipline. A badly designed test produces a confident wrong answer, which is worse than no test.

You're at the scale where modelling pays. MMM needs data volume and expertise. Below meaningful spend it's premature; above it, it's frequently the most durable measurement you can buy — MMM and revenue ops specialists.

Channel execution is the real problem. Sometimes the numbers are fine and the campaigns aren't — paid social or Google Ads specialists.

When vetting: ask how they'd handle platforms reporting more revenue than Shopify recorded. Anyone claiming they can reconcile it exactly is overselling. The right answer involves blended metrics and testing.

Add up what your platforms claim and compare it to Shopify. That comparison ends the reconciliation debate faster than any meeting.

Then set blended performance as the number you make budget decisions on, add a post-purchase survey this week because it costs almost nothing, and plan one holdout test for the channel you're least sure about. That's a working measurement setup, and it's achievable without new software.

If your tracking is genuinely broken rather than just imprecise, that's worth fixing first — browse marketing analytics and attribution experts, or MMM specialists if you're at the scale where modelling earns its cost. Free matching, verified experience, no commissions.

Related guides on measurement, growth and profitability:

• Marketing Analytics & Attribution Experts — tracking setup, blended reporting and testing design
• Marketing Mix Modelling Experts — privacy-durable measurement at scale
• Retention & Lifecycle Marketing — cohort measurement and lifetime value
• Shopify CRO — the testing discipline incrementality work borrows from
• Ecommerce Accounting on Shopify — getting the margin data attribution decisions depend on
• Best Shopify Apps 2026 — the analytics category alongside the rest of your stack
• Signs You've Outgrown Your Shopify Setup — when reporting gaps are a symptom of something larger