Your Meta dashboard says the campaign returned 3.8x. Triple Whale says 2.1x. GA4 attributes a third of the revenue to direct. Your CFO wants one number by Friday. Welcome to ecommerce attribution in 2026, where every platform is technically right and operationally useless.
The problem is not that the tools are broken. The problem is that most teams read attribution data the way they read horoscopes: scanning for the bit that confirms what they already wanted to do. After a decade of running paid budgets for subscription and high-LTV Shopify brands, here is the working rule set we use to read these dashboards without losing the plot.
1. Treat ecommerce attribution as a triangulation problem, not a truth problem
No platform sees the full customer journey. Meta sees what happened inside Meta. Google sees what happened inside Google. GA4 sees what its consent banner lets it see. Triple Whale stitches click and post-purchase survey data into one view but still relies on the same upstream signals. None of them is lying. None of them is complete.
Once you accept that, the question shifts from which dashboard is right to where do the dashboards agree, and where do they disagree. Agreement is signal. Disagreement is where the interesting decisions live.
2. Know the difference between attributed and incremental revenue
Attributed revenue is the revenue a platform claims credit for. Incremental revenue is the revenue you would not have earned without that channel running. They are not the same number, and the gap between them is where most DTC brands quietly waste budget.
Brand search is the classic example. Google Ads will happily report a 12x ROAS on your brand campaign. A geo holdout test usually shows that 70-90% of that revenue would have arrived organically anyway. The platform did not lie. The reporting just answered a different question than the one you needed answered.
"Attributed revenue tells you what a channel touched. Incremental revenue tells you what a channel caused. Only one of those pays your salary."
3. Accept that GA4, Meta and Triple Whale will never reconcile
Stop trying to make them agree. They use different models, different attribution windows, different click and view definitions, different consent handling, and different ways of treating direct traffic. A 20-40% spread between platforms for the same week is normal. A 5% spread usually means something is wrong with your data layer, not that the universe has aligned.
What you can do is pick one source as your source of decisions. Most of the brands we work with pick Triple Whale or Northbeam for daily channel decisions, GA4 for on-site behaviour and content questions, and Shopify analytics for the cash-in-bank reality check. Each tool has one job. None of them is asked to do all three.
CRO Obsessed
Stop arguing with your dashboards. Start asking them better questions.
PMD is a full-funnel CRO and profit-optimisation agency. We help subscription and high-LTV Shopify brands fix the funnel that sits between their ad spend and their profit.
Book a 30-min call with Paddy McLarnon →4. Use MMM and MTA for different questions, in plain English
Multi-touch attribution (MTA) is bottom-up. It tries to credit each touchpoint a user clicked on the way to purchase. It is good at answering which campaign in this channel is working. It is bad at measuring channels it cannot see, which now includes a lot of paid social on iOS.
Marketing mix modelling (MMM) is top-down. It looks at spend by channel over time, controls for seasonality and external factors, and statistically estimates each channel's contribution. It is good at answering where should the next pound of budget go. It is bad at telling you which ad creative inside Meta is the winner.
Use MTA inside platforms for creative and campaign decisions. Use MMM across platforms for budget allocation decisions. If your team is using MTA to set monthly budget splits between Meta and Google, you are using the wrong tool for the job.
The two-question test
- Which creative should we kill? MTA inside the platform.
- How much should we spend on Meta vs Google next quarter? MMM, ideally validated with holdout tests.
5. Replace ROAS with three sharper questions
ROAS is a vanity metric pretending to be a financial one. A 4x ROAS on a product with 20% contribution margin loses you money. A 1.8x ROAS on a subscription product with a £180 LTV is a printing press. ROAS without context is noise.
The three questions that actually matter:
- What is our CAC ceiling? The maximum cost per new customer the unit economics can absorb, given your contribution margin and target payback window. If you have not modelled your LTV-to-CAC ratio properly, no attribution dashboard will save you.
- What is our marginal CPA? Not your blended CPA. The cost of the next customer at the next pound of spend. Average CPA hides the fact that scaling spend usually raises CAC.
- What is the post-purchase LTV by acquisition source? Customers acquired on a discount code from a deal-site partner behave nothing like customers acquired from a brand-led YouTube ad. Same CAC, very different value over 12 months.
6. Know when to trust the platform claiming credit, and when to ignore it
Some claims are more credible than others. A platform showing a 1-day-click conversion on a £200 product to a first-time buyer is probably a real, causal touchpoint. A platform showing a 7-day-view conversion on a repeat customer who also got two emails and saw your brand on TikTok is almost certainly piggybacking on demand someone else created.
A working heuristic: weight click-based attribution more heavily than view-based. Weight first-time-buyer conversions more heavily than repeat. Discount view-through claims on customers who were already in your email list. The platforms will not do this for you. Your team has to layer the judgement on top.
7. Make decisions with conflicting data using a decision hierarchy
You are never going to have clean data. You can still make clean decisions if you agree the hierarchy in advance:
- Holdout test results trump everything. If you turned a channel off for a region or two weeks and revenue moved, that is the closest thing to truth you will get.
- Blended efficiency over a 30-day window is the next most reliable signal. Total ad spend divided by total new-customer revenue, ignoring platform claims.
- MMM outputs are useful for quarterly budget shifts. Not for tomorrow's bid changes.
- Platform ROAS is the noisiest, most biased input. Use it for within-platform decisions only.
When platforms disagree about which channel deserves credit, your default move is to look at blended efficiency over time and ask: did the trend change when we changed the spend? That question is harder to lie about.
8. Spot the warning signs that your reporting is being gamed
Sometimes the data is being shaped by the people presenting it. Things to watch for in agency or in-house reporting:
- Attribution window quietly extended. Last month's report ran on 7-day click. This month's mysteriously uses 7-day click plus 1-day view. ROAS goes up. Nothing else changed.
- Brand and non-brand search blended into one ROAS line. Brand search props up the number while non-brand quietly bleeds money.
- Reporting compares paid-only revenue to total ad spend. The denominator includes channels not in the numerator. The ratio looks better than reality.
- Top-line revenue cited, contribution margin hidden. Revenue can grow while profit collapses, especially if discount-led spend is doing the lifting.
- Cohort LTV stops being shared. When acquisition quality drops, LTV trends are the first thing that gets dropped from the deck.
If any of these patterns shows up in your reporting, the question is not whether your attribution is wrong. The question is whether someone is using its complexity as cover. The same scrutiny we apply in a Shopify CRO audit applies to the reporting deck itself.
9. Run small holdout tests to keep the system honest
Every quarter, run one structured holdout. Turn off a single channel in a single geography for two weeks. Measure the revenue delta against a matched control geography. The result will rarely match what the platform said that channel was contributing.
Holdouts are uncomfortable because they cost short-term revenue. They are also the only way to keep your attribution honest. Brands that run two or three holdouts a year tend to find that one of their channels is contributing far less than the dashboard suggests, and they redirect that spend into channels that move the blended number. That redirect is where the real profit lives, alongside the on-site work covered in our Shopify profit optimisation framework.
This is also where the case for a proper profit optimisation programme tends to land hardest: once you can see what is actually driving revenue, the next obvious question is what is leaving money on the table inside the funnel itself.
FAQs
Why don't Meta, Triple Whale and GA4 agree on revenue?
Because they answer different questions with different data. Meta sees its own click and view data within its window. GA4 relies on consented cookies and a different attribution model. Triple Whale stitches in post-purchase survey data. A 20-40% spread is normal. Pick one as your source of decisions and stop trying to reconcile them.
Should we use MMM or MTA for ecommerce attribution?
Both, for different jobs. Use MTA inside each platform for creative and campaign decisions. Use MMM across platforms for quarterly budget allocation. Validate MMM outputs with holdout tests when the stakes are high.
What is incremental ROAS and how is it different from platform ROAS?
Incremental ROAS measures the revenue that would not exist without the channel running. Platform ROAS measures the revenue the platform claims to have influenced. The gap is biggest on brand search, retargeting and view-through-heavy campaigns.
How often should we run holdout tests?
At least once a quarter, on a different channel each time. Two-week minimum, geo-based where possible. Yes, it costs short-term revenue. It is still the cheapest insurance you can buy against misallocating budget for a full year.
Is GA4 still useful for ecommerce attribution?
Yes, but not as your primary attribution source. GA4 is excellent for on-site behaviour, content performance and funnel diagnostics. It is a weaker tool for cross-channel attribution because of consent loss and modelling assumptions. Use it for what it is good at.
Who should own attribution decisions inside a DTC team?
Ideally a head of growth or CFO-aligned analyst, not the platform managers. Anyone responsible for spending on a channel has an incentive to read that channel's data generously. Whoever owns attribution should own no individual channel budget.
Full-funnel CRO. Profit obsessed.
Want this on your store?
We help subscription and high-LTV Shopify brands turn cold traffic into post-click profit. Strategy, copy, design, development, and CRO under one roof.