Shopify says one number. GA4 says another. Meta and Google can each claim enough revenue that, added together, they exceed what the company actually sold. That doesn’t automatically mean the tracking is broken, and it doesn’t mean anyone is lying. It means four systems are answering four different questions.
This article is for owners who get reports from all of them and want to know which number to use for which decision — without becoming an attribution specialist.
None of them is the single correct number, because they don’t measure the same thing. Believe each one for the question it was built to answer.
- What the business soldYour store or first-party order data, reconciled with your accounting records.
- What each ad platform credits to its adsMeta and Google Ads, mainly for running and optimizing their own campaigns.
- How people found and used the siteGA4, a cross-channel analytics view with its own gaps.
- What marketing actually causedCompany economics for most decisions, and controlled tests when the decision is big enough to justify them.
In this article
Why don’t Meta, Google, Shopify and GA4 agree on revenue?
Because each is built to answer a different question. The useful question isn’t “which dashboard is lying?” It’s “which system should govern this decision?”
-
Question 1
What did the business actually sell?
- Primary source
- First-party order data, reconciled with accounting records
- Good for
- Orders, net sales, refunds, new customers
-
Question 2
What does each ad platform think its ads influenced?
- Primary source
- Meta Ads Manager, Google Ads
- Good for
- Bidding, campaign and creative decisions inside each platform
-
Question 3
How did people move through our site?
- Primary source
- GA4, with store data for context
- Good for
- Traffic sources, behavior, cross-channel paths
-
Question 4 · the hardest
How much revenue did marketing actually cause?
- Primary source
- Company economics, and controlled experiments when the decision justifies them
- Good for
- Budget decisions: what would have happened without the ads?
The first three questions each have a dashboard. The fourth doesn’t. No ordinary attribution report proves what would have happened if the advertising hadn’t run, and that is the question most budget decisions actually turn on.
What does each system see when one customer places one order?
One real event, observed from four positions, plus a fifth question none of them answers on its own.
- Store / order data
“One order occurred, for this amount, to this customer.”
Records the transaction itself - GA4
“Here is how analytics classified the journey that led here.”
Credits channels by its own attribution settings - Google Ads
“A Google Ads interaction preceded this, within our window.”
Credits value by the account’s conversion settings - Meta
“A Meta ad interaction preceded this, within our setting.”
Credits value by the ad set’s attribution setting
Which marketing activity changed what would otherwise have happened?
Not visible in any of the four views aboveEach view is accurate within its own rules. The mistake is reading them as four separate sales, or treating any one of them as the answer to the incrementality question underneath.
What is each system actually measuring?
Each tool has a role. None is omniscient.
Your store and order data
Your commerce platform records transactions: orders, gross sales, discounts, refunds, net sales and customers, depending on the report. Shopify, for example, defines net sales as gross sales minus returns, allowances and discounts, and its customer reports separate first-time from returning customers. That makes it the right starting point for store transactions. It isn’t automatically the company’s formal revenue figure. Your accounting records may differ because of reporting periods, taxes, shipping, chargebacks, cancellations, other sales channels and revenue-recognition policy. For store transactions, start with first-party order data. For formal company revenue, reconcile it with your books.
GA4
Google Analytics tries to observe and organize sessions, users, traffic sources, events and purchases, then assign credit. Its reporting attribution model is a property setting, and changing it rewrites historical as well as future data in the affected reports. It also depends on what it can observe. Google’s documentation says that when visitors decline consent, Analytics is missing their data; behavioral modeling can estimate some user and session metrics for eligible properties. Browser restrictions, blocked scripts and people switching devices create further gaps. GA4 is valuable. It isn’t a complete record of what you sold.
Google Ads
Google Ads reports the conversions and value it attributes to Google Ads interactions under each conversion action’s settings. Two settings matter most to an owner. The conversion window is how long after an ad interaction a conversion can still be recorded; it’s set per conversion action, and changes apply only going forward. The attribution model decides how credit is split: Google now supports data-driven and last-click attribution, has retired first-click, linear, time-decay and position-based models, and uses data-driven as the default for most conversion actions. Google Ads also dates a conversion to the ad click rather than the purchase in its standard reports, one of the reasons it and Google Analytics can disagree even with a correct setup.
Meta
Meta reports conversions and value it attributes to Meta ads under the attribution setting chosen for each ad set. Meta’s help documentation describes crediting a purchase to an ad when it happens within a set number of days after someone viewed or clicked it. Those rules changed recently: in March 2026, Meta announced that for website and in-store conversions, click-through attribution now counts link clicks only, with likes, shares, saves and qualifying video views reported separately as “engage-through,” and billing unchanged. Settings and definitions like these change, so check what your account uses today rather than relying on any article’s description, this one included.
Platform attribution isn’t worthless because it’s attributed. Meta and Google need those signals to optimize delivery and bidding. The mistake is reading “Meta attributed $300,000” as “Meta independently created $300,000 that wouldn’t otherwise exist.” Those are different claims.
Can Meta and Google both count the same sale?
Yes. When a customer touches both platforms before buying, each can credit the purchase to its own ads. That’s different lenses on one journey, not fraud.
- MondayClicks a Meta prospecting adBrowses, doesn’t buy
- WednesdayOpens a brand emailDoesn’t click through
- FridaySearches the brand on GoogleClicks the paid brand ad
- FridayBuys a $150 orderOne transaction
Depending on configuration and eligible interactions, who may report it
- Store$150Records one order
- Metaup to $150If Monday’s link click falls inside the ad set’s click window
- Google Adsup to $150Friday’s brand-search click precedes the purchase
- Email platformup to $150If its own rules credit an opened email
- GA4$150, split or assignedDistributes credit by the property’s attribution model
Credit reported: up to $450 across Meta, Google Ads and email.
Orders that happened: one, worth $150.
Nobody has to be wrong for this to happen. Each platform can see its own interactions and uses its own rules. What can’t happen is adding their claims together and calling the total company revenue.
Why can attributed sales add up to more than total revenue?
Because the same orders receive credit in more than one system. Here’s one illustrative month.
- Store net salesWhat the business sold
$500,000
- GA4 purchase revenueWhat analytics observed
$445,000
- Meta attributed revenueCredited to Meta ads
$310,000
- Google Ads attributed revenueCredited to Google ads
$270,000
- Meta + Google, added togetherNot a real revenue figure
$580,000
The vertical line marks actual net sales ($500,000). Bars share one scale.
- Meta + Google claims
- $580,000, or 116% of net sales: $80,000 more than the business sold
- GA4 vs. store
- $445,000, or 89% of net sales: $55,000 below the store, for any of the reasons in the next section
- Platform ROAS
- Meta 4.4x on $70,000 of spend; Google 6.0x on $45,000
- Company view
- $500,000 ÷ $115,000 of ad spend = 4.3x MER; $115,000 ÷ 2,300 new customers = $50 blended CAC
The company didn’t secretly sell $580,000. Some orders were credited by both platforms. And GA4 sitting below the store isn’t proof of a problem either; the next section covers why that happens. For the executive view, the store figure, the new-customer count and the blended numbers built on them are the ones that describe the business. Why a strong-looking ROAS can still coexist with weak profit follows the same month-level logic into margins.
Why does GA4 revenue not match Shopify?
Usually for one of five kinds of reason. Only the last one means something is broken.
- 01Definition
They may not be measuring the same “revenue.” Taxes, shipping, discounts, refunds and cancellations can be included in one and not the other.
- 02Collection
One system recorded activity the other didn’t observe: declined consent, blocked scripts, browser restrictions, orders placed outside the website.
- 03Attribution
They assign credit differently, by model, window and which channels are eligible.
- 04Timing
The same purchase lands in different periods: reporting time zones, click date versus purchase date, processing delays.
- 05Implementation
The tracking is actually broken: purchase events missing, firing twice or failing after a site change.
Most month-to-month differences come from the first four. They’re worth understanding, not fixing. The fifth is worth fixing quickly, because every downstream number inherits it.
How do windows, views, devices and brand search change the numbers?
Each one moves attributed revenue without any change in what customers actually did.
Attribution windows
A window answers one question: how long after an eligible marketing interaction can a later purchase still receive credit? Meta sets it per ad set; Google Ads sets it per conversion action; GA4 has its own lookback settings. Longer windows credit more purchases; shorter ones fewer. Two platforms with different windows will report different revenue from the same customers. Defaults and options change, so ask your team which windows each account uses now.
View-through and engagement credit
Some systems can credit a purchase after someone saw or engaged with an ad without clicking through. That’s useful information: the platform observed an exposure, then a purchase, inside its measurement rules. But a view followed by a purchase doesn’t, by itself, prove the view caused the purchase.
Devices and identity
A customer may see an ad on a phone, research on a laptop and buy on a tablet. Each system connects those steps with different success. The owner-level lesson: perfect one-to-one journey reconstruction is often unrealistic, so treat channel paths as evidence, not a transcript.
Brand search
Branded search often sits at the end of journeys that began somewhere else. A Meta ad creates interest; the customer later searches your name and clicks a Google ad. Google Ads may claim the order, Meta may too, and the order may have needed both, one or neither. Field experiments at eBay found that its paid search on its own brand name had no measurable short-term benefit, though eBay is an unusually well-known brand and the finding doesn’t transfer automatically. Brand search can still protect your name from competitors and capture intent efficiently. The point is that it complicates channel-level credit, not that it’s worthless.
Returning customers
A returning customer who clicks an ad and buys again is real revenue. It just isn’t new-customer acquisition. So separate two questions: how much revenue did ads touch, and how many genuinely new customers did the business add? Divide acquisition spend by actual new customers and you get the number that tells you what growth costs, which first-order break-even CAC then tells you whether you can afford.
What’s the difference between attribution and incrementality?
Attribution assigns credit. Incrementality asks what the marketing actually caused.
Answers: which touchpoints get credit under these rules?
Example: a long-time customer sees a Meta retargeting ad, later searches your brand on Google and buys. Meta may claim it. Google may claim it. GA4 assigns it by its model. The store records one order.
Answers: would this customer have bought without either ad?
For a loyal customer searching your name, possibly yes. None of the attribution reports can say, because answering requires a comparison against a world without the ads.
That comparison is what controlled experiments provide. Google describes its Conversion Lift studies as splitting an audience into a group that sees the ads and a group that doesn’t, and measuring the difference as the lift the ads caused; the tool isn’t available to every account. Geo experiments and holdouts work on the same principle. Research using large Facebook experiments found that the observational methods commonly used in the industry often failed to reproduce what the randomized experiments showed.
Experiments aren’t easy either. Across 25 large field experiments with major U.S. retailers and brokerages, economists found that the median confidence interval on return on investment was more than 100 percentage points wide. Sales are noisy relative to what an ad costs, so informative tests can need very large audiences. That’s why the right level of rigor depends on the decision. Not every company needs an experiment every month.
Which number should govern which decision?
Match the system to the question, and never let one view stand alone for the decisions that matter most.
| The decision | Primary view | Supporting views | Don’t use alone |
|---|---|---|---|
| How much did the company actually sell? | First-party order data, reconciled with accounting | GA4 for traffic and behavior context | Meta + Google attributed revenue |
| How should Google optimize Google campaigns? | Google Ads conversion data, if implementation is sound | Business economics, new-customer quality | Store totals, which can’t steer bids |
| How should Meta optimize Meta campaigns? | Meta conversion and event data, if implementation is sound | Business economics, new-customer quality | Store totals, which can’t steer delivery |
| How are people finding and using the site? | GA4 | First-party order data for context | Any single platform’s view of its own channel |
| What does it cost to add a new customer? | Acquisition spend ÷ actual identifiable new customers | Platform reports for channel diagnosis | Spend ÷ platform-reported conversions |
| Should we increase the total marketing budget? | Company economics: new-customer CAC, contribution, conversion, retention, capacity, cash, measurement confidence | Platform trends | Platform ROAS |
| Did the ads cause the additional sales? | Incrementality evidence suited to the decision | Company-level trends before and after changes | Ordinary attribution reports |
Should rigor rise with the size of the decision?
Yes. An owner doesn’t need perfect attribution before every decision. But the cost of a measurement error grows as spend increases, as CAC approaches what a customer can support, as the platforms disagree more, as first-order margins thin, and as the business considers a large scale-up.
- Routine optimizationShifting budget between ads or campaigns
Platform data, sanity-checked against company trends
- Material budget changeMoving meaningful money between channels
Company-level new customers, CAC and contribution; reconciled reports
- Major scale-upCommitting substantially more to demand
All of the above, plus the strongest causal evidence the decision justifies
The amount of measurement rigor should rise with the consequence of the decision.
Should I use MER instead of ROAS?
Use both. MER is a useful executive check, not an attribution solution.
A blended ratio, total revenue divided by total ad spend, avoids choosing which platform gets credit, which makes it good for watching the trend at company level. But it doesn’t subtract product or order costs, it doesn’t prove what the ads caused, it can improve because returning customers or organic demand grew, and it can worsen because the business deliberately invested ahead of revenue. In the example above, MER was 4.3x. That says nothing yet about profit or about what the ads caused. Read it with new customers, CAC and contribution, the numbers in the monthly owner scorecard.
When are differences normal, and when should I worry about tracking?
Consistent, explainable gaps are normal. Sudden, unexplained or impossible ones deserve investigation.
- Ad platforms credit overlapping revenue
- Analytics and ad platforms use different attribution rules
- Minor timing and reporting-period differences
- Returning customers touch several channels
- Cross-device journeys are only partly observed
- GA4 purchases far below store orders for a sustained period, with no understood reason
- Duplicate transaction IDs, or purchase events firing more than once
- Purchase events failing entirely
- Tracking changing suddenly after a site update
- A platform reporting impossible values
- Large, unexplained breaks in a trend
There’s no universal percentage that separates the two; the test is whether the gap is stable and explained. One specific safeguard is worth knowing about: Google Analytics de-duplicates purchases that share a transaction ID, and warns that sending the same ID for different orders can significantly undercount them. You don’t need to check that yourself. You need someone who can tell you it’s right.
A short tracking health check
- Does the store reliably record every order?
- Can we identify actual new customers?
- Do unique transaction IDs prevent duplicate purchases?
- Did tracking change after a recent site update?
- Are unexplained gaps getting larger?
- Can the team explain why the major systems differ?
- Do company-level sales and acquisition trends roughly support the channel reports?
What should I ask my agency or marketing team?
Nine questions, asked constructively. Good teams will welcome them.
- What number are we using for actual company sales?
- How are we defining a new customer?
- Which numbers in this report are platform-attributed?
- Are Meta and Google both claiming some of the same orders?
- What is our blended new-customer CAC, using actual new customers?
- How much paid revenue came from returning customers?
- What changed in tracking this month?
- How confident are we in the measurement behind the recommendation to spend more?
- What evidence would change your recommendation?
Clear answers are a good sign about the people giving them. If the answers stay unclear, how to tell whether your ad account is actually being managed is a sensible next read.
How accurate does measurement need to be before I increase ad spend?
Accurate enough that the company-level story is coherent. High confidence doesn’t mean every dashboard matches.
- First-party orders are reliable
- Actual new customers can be identified reasonably well
- Marketing spend is known
- Major tracking is functioning
- Differences between systems are understood well enough
- Company economics and marketing trends tell a coherent story
- Nobody can reconcile orders
- New vs. returning customers is unknown
- Purchase events appear broken
- Tracking changed recently
- Major systems disagree for unexplained reasons
- The company is scaling mostly on attributed ROAS
Low confidence doesn’t forbid decisions; it should reduce confidence in a material scale decision. That’s why measurement confidence sits across every layer when you’re working out whether marketing is really your growth bottleneck: if the numbers can’t be trusted, neither can the diagnosis built on them.
So which one should I believe?
Each one, for its own question. Your store and books for what the business sold. Meta and Google for running their own campaigns. GA4 for how people found and used the site. And for what marketing actually caused, company economics most of the time, with more rigorous evidence as the decision grows.
The discipline is refusing two shortcuts: adding the platforms’ claims together and calling it revenue, and picking whichever dashboard shows the biggest number. Neither describes the business.
One transaction, several measurement systems, different perspectives. Decide which one each decision needs.
Frequently asked questions
Why does Shopify revenue not match GA4?
They measure different things in different ways. Shopify records store transactions; GA4 observes browsing and purchase events it can collect and attributes them by its settings. Differences in revenue definitions, consent and blocked scripts, attribution, reporting periods, or genuinely broken purchase tracking can all create a gap.
Why does Meta show more revenue than Shopify?
Meta reports the value it credits to its ads under the ad set’s attribution setting. Some of those orders are also credited by Google or other channels, and some customers may have bought anyway. Meta’s figure is attributed revenue, not a count of unique sales.
Can Meta and Google both count the same sale?
Yes. If a customer interacted with both platforms’ ads before buying, each can credit the purchase under its own rules. That’s why their attributed revenue can add up to more than total sales.
Which platform should be my source of truth for sales?
For store transactions, your commerce platform or first-party order data. For formal company revenue, your accounting records, reconciled with the store. Ad platforms and analytics tools describe marketing activity, not total sales.
Is Shopify more accurate than GA4?
For counting store orders, first-party order data is usually more complete, because it records the transaction itself. GA4 is more useful for understanding traffic sources and behavior. They’re accurate at different jobs.
Should I trust Meta’s ROAS?
Trust it as Meta’s view of the revenue its ads touched, which is useful for managing Meta campaigns. Don’t treat it as proof of what Meta caused, or as a measure of profit. For budget decisions, compare it with company sales, new customers, CAC and contribution.
Which revenue number should I use for ROAS?
Inside a platform, its own attributed revenue, so the comparison is consistent. At the company level, use net sales against total ad spend (MER), knowing it doesn’t prove causation or subtract costs.
What is the difference between attribution and incrementality?
Attribution assigns credit for a conversion to marketing interactions under a set of rules. Incrementality measures the additional sales the marketing caused compared with what would have happened without it, usually through a controlled test.
How accurate does attribution need to be before I increase ad spend?
It doesn’t need to be perfect. It needs to be good enough that store sales, new customers, CAC and marketing trends tell a coherent story. The larger the proposed increase, the more confidence it deserves before you act.
Why does Google Ads revenue not match GA4?
They can use different attribution settings and eligible channels, and Google Ads dates conversions to the ad click in its standard reports. Google documents that differences between the two are common even when both are set up correctly.
Sources & further reading
- Meta rewrites click attribution rulesPPC Land, reporting Meta’s March 3, 2026 announcement.
- About actions attributed to your adMeta Business Help Center.
- About conversion windowsGoogle Ads Help.
- About attribution modelsGoogle Ads Help.
- Conversion discrepancies between Google Analytics and Google AdsGoogle Ads Help.
- About Conversion LiftGoogle Ads Help.
- Select attribution settingsGoogle Analytics Help.
- Behavioral modeling for consent modeGoogle Analytics Help.
- Minimize duplicate key events with transaction IDsGoogle Analytics Help.
- Customers reports and What Are Net Sales?Shopify.
- A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at FacebookGordon, Zettelmeyer, Bhargava and Chapsky, Marketing Science, 2019.
- Consumer Heterogeneity and Paid Search Effectiveness and The Unfavorable Economics of Measuring the Returns to AdvertisingBlake, Nosko and Tadelis, Econometrica, 2015; Lewis and Rao, Quarterly Journal of Economics, 2015.
Platform behavior described here reflects documentation available in September 2026 and can change; check your own account settings. The journey, the order and the monthly figures are illustrative, created to show the reasoning. They aren’t benchmarks or client data.
