Your agency can improve reported Meta ROAS while your business keeps less money. The purchase event records the checkout; it does not wait to see whether the order comes back.
That is the fact I would put at the top of every ecommerce performance report: a purchase event confirms an order, not revenue you kept. Until the return window closes and the costs come out, the calculation is unfinished.
This is not an obscure edge case. NRF and Happy Returns estimate that 19.3% of online sales came back in 2025. Say you sell two products at $60 each, both showing a 2.4x ROAS in Ads Manager. One comes back a third of the time. The other almost never does. The dashboard gives them the same score. Your bank account does not.
Free tool: the ROAS Calculator works out your return on ad spend, your break-even ROAS and the budget a revenue goal takes.
I’ll build the calculation from checkout to retained revenue, then contribution. The useful question is not whether your agency’s report is wrong. It is whether the information missing from it would change where you put the next dollar.
A recorded purchase starts the calculation
A purchase event gives an account manager a timestamp, currency and order value to work with. Alongside attribution and spend, those signals help them assess which campaigns produce orders and what those orders cost. That is useful work. Purchase tracking beats counting clicks, and it beats guessing.
But the event records what happened at checkout, not what settled afterward. Meta generally does not retroactively subtract refunds from historical campaign revenue. If a customer returns the product two weeks later, the original report does not change with it.
Here is the arithmetic. In a worked example from Appnary:
- A campaign spends $1,000 and produces 20 orders at $60.
- Ads Manager reports $1,200 in purchase revenue: 1.2x ROAS.
- Six orders come back over the next three weeks, producing $360 in refunds.
- Retained revenue falls to $840: 0.84x revenue after returns, against the same spend.
The dashboard still says 1.2x. The purchase events happened. So did the refunds.
Even that second ratio is not profit. Product costs, fulfillment and payment fees still have to come out. I use retained revenue here to mean revenue after refunds and cancellations, not money left after every cost. Keep those definitions separate or the next calculation becomes another impressive-looking number answering the wrong question.
Same checkout ROAS, different money left
Take two illustrative campaigns and hold the starting conditions steady. Each spends $10,000 a month. Each reports $20,000 in attributed purchase revenue. Ads Manager gives both a 2.0x ROAS.
Now apply their return rates:
| Measure | Campaign A | Campaign B |
|---|---|---|
| Ad spend | $10,000 | $10,000 |
| Reported purchase revenue | $20,000 | $20,000 |
| Reported ROAS | 2.0x | 2.0x |
| Return rate | 8% | 32% |
| Revenue after returns | $18,400 | $13,600 |
| Revenue after returns ÷ ad spend | 1.84x | 1.36x |
For this example, assume the return rates translate into the same proportions of refunded revenue. That keeps the calculation focused on the gap rather than differences in order value.
Campaign A retains about 35% more revenue than Campaign B. The campaigns tie at checkout and separate after the return window. Neither retained-revenue figure tells us profit yet, but we already have a reason not to treat the campaigns as interchangeable.
Scaling multiplies the missing information
An agency looking only at the reported ratio has no reason to prefer A over B. If B starts generating more checkout revenue for the same spend, its reported ROAS improves and it becomes the obvious candidate for more budget.
Except the returns have not disappeared. More sales of the same refund-heavy product bring more refund exposure. If budget shifts away from a product that retains more revenue, the account can improve its reported ROAS while weakening its economics.
That is the mechanism. The report rewards purchase value; the business lives on what survives the purchase.
Product costs add another layer. A campaign selling a high-revenue, low-margin product can be less attractive than one with a lower reported ROAS and a healthier product mix. I would not scale either campaign from the table alone. I would first ask what it costs to deliver and keep those sales.
Revenue after returns still is not contribution
The next layer is a reconciliation, not a new attribution model. Start with the orders and work down:
- Subtract discounts, returns and cancellations from gross sales to reach net sales.
- Subtract landed product costs to understand gross profit.
- Account for relevant variable costs, including processing and fulfillment.
- Subtract ad spend to assess contribution after advertising.
The labels matter less than consistent treatment. Write down which costs your contribution figure includes. Otherwise, marketing and finance can spend a meeting agreeing on a word while discussing different numbers.
For Campaigns A and B, this means attaching cost assumptions to the revenue that survived. A’s lower return rate is useful information, but it does not prove A is more profitable if its costs differ materially from B’s.
This is where the product mix earns its place in the budget discussion. The ROAS column reports attributed purchase value. It cannot, by itself, tell you which mix leaves the business better off.
Two feedback loops, not one magic tracking fix

I separate this into a campaign loop and a business loop. They inform each other, but they cannot run on the same clock.
The campaign loop needs usable value signals
The campaign loop covers bidding, budgets, creative rotation and audience expansion. It acts on the signals configured and available to the platform.
Meta documents value optimization for profit, and its guidance emphasizes consistent conversion values that reflect differences in business outcomes. The useful principle is straightforward: a checkout total and a profit-informed value are not the same objective.
But a margin signal is not automatically a returns signal. Product margin alone does not tell the system which orders will come back. If the value you send ignores a meaningful difference in expected refunds, that difference remains missing.
Before changing the signal, have the implementation checked against the documented setup. Agree on what the value represents, how consistently it can be supplied and which costs or return assumptions it includes. Renaming revenue “profit” does not improve the input. It just gives the mistake a better job title.
The business loop decides what deserves more budget
The slower loop answers questions outside the platform:
- Should Campaign B keep its budget when its returns are four times A’s?
- Does a campaign that clears the ROAS target still contribute enough after costs?
- Is the biggest refund reason something the merchant can fix?
- Would a different product mix retain more of the revenue already being acquired?
These decisions need reconciled order economics, not perfect attribution. Product-level retained revenue, credible cost assumptions and actionable refund reasons can change the brief even when they cannot rewrite the platform’s historical report.
For A and B, that means reviewing both campaigns after their orders have had time to return. Then decide whether to change budgets, fix the product experience or revise the values used for optimization. The answer need not be “turn B off.” It does need to be more informed than “both show 2.0x.”
Let each number do its job
Refunds arrive later than purchases. Platform attribution and store accounting also answer different questions, so their totals will not always match.
I would not spend a quarter chasing agreement to the cent. Use platform reporting to understand delivery and reconciled store data to judge economics. Investigate a widening or unexplained gap; do not turn every predictable difference into a forensic project.
The goal is a better budget decision, not a variance report with excellent formatting.
Give your strategist a table, not another dashboard
Before the next strategy call, assemble one table covering at least a full quarter, so most orders have had time to pass through the return window. Use SKUs or product groups as rows.
Include:
- Gross revenue, refunds and cancellations.
- Net revenue and return rate.
- Landed product cost.
- Relevant processing and fulfillment costs.
- Contribution, with its calculation stated.
The table should rank products by what the business keeps, not just what customers initially buy. That is the missing view in our A-versus-B comparison.
Add refund reasons in categories someone can act on. Fit and sizing may point toward product pages or a size guide. Damage may point toward packaging or the supplier. Late delivery may point toward shipping expectations. A refund count tells your strategist there is a problem; a reason helps identify the work.
Then make the cost assumptions explicit. Include freight and duties in landed cost where relevant, and account for discount stacking. If your estimated margin sits between 38% and 44%, say so. A stated range is more useful than a precise number copied from a two-year-old spreadsheet nobody trusts.
Keep personal information out of the strategic brief
This analysis does not require a customer list. Aggregate product economics are enough for the strategic review: return rates by product, refund reasons grouped into categories and contribution by SKU or margin band.
Names, emails and shipping details do not help answer whether Campaign B deserves more budget. Do not circulate them with the table.
Sending conversion signals to Meta is a separate process. Use the documented implementation and applicable consent requirements. Keep that tracking pipeline distinct from the aggregate business analysis your strategist needs.
Change the brief before changing the agency
The first response to a refund gap is often to start comparing agencies. I would first test whether the missing information changes a decision.
Suppose every campaign’s reported purchase revenue is overstated by the same 15%. Their ranking stays the same, but that does not mean the error is harmless. Campaigns can still fall below the business’s acceptable return, and the right total budget can change.
Now apply product-specific return rates. If the campaign you are scaling falls behind one you are cutting, the problem reaches beyond the target: it changes the allocation.
Ask about thresholds and rankings, not just the revised ROAS number. Which campaigns still clear the bar? Which deserve the next dollar? Those are different questions, and your brief should ask both.
For the tracking baseline behind that conversation, I would start with making purchase numbers agree before hiring a Meta ROAS agency. Bring the baseline and the retained-revenue table to your current strategist first. Fire second, if at all.
Add one monthly review with a clear question
Alongside the platform report, ask for a short retained-revenue review: the same campaigns and order period, with refunds and cancellations accounted for, plus contribution where costs are available.
Allow for the return window rather than treating the newest orders as settled. It need not reconcile to the cent. It needs to answer: did the campaign ranking survive contact with the returns, and do the campaigns still clear our economic bar?
That is a reporting requirement someone can act on. “Show us better ROAS” is not.
What an accountable operating partner does differently
This is where the operating model matters. Business context has to change execution, not sit in a strategy deck.
At groas, universal business context means execution is informed by margins, sales cycles and goals rather than purchase events alone. Actions carry plain-language reasoning, and a named senior strategist owns the outcome. Its flat monthly fee also removes the percentage-of-spend incentive to earn more simply because the media budget grows.
That is a stronger accountability model than paying an agency 10% of spend while judging it only on checkout ROAS. The percentage fee rewards more spend; the reporting target rewards more attributed purchase value. Neither, on its own, rewards revenue surviving the return window.
I would judge the partner by what changes once the missing economics arrive. Can they explain the budget decision? Can they identify the assumption behind it? Can you see what they did? groas’s results page shows its client-verified outcomes across paid and organic search. Those are the standards I would bring to this conversation, not another promise to watch the dashboard more closely.
Run a fresh case before the next budget decision
Take an illustrative merchant spending $30,000 a month across four Meta campaigns. Together, they report $90,000 in purchase revenue: 3.0x ROAS. The store’s quarterly refund rate is 22%, concentrated in the leading campaign’s fit-sensitive product, which has a 38% return rate. The other three sit near 10%.
The account average tells us there is a gap. It does not tell us which campaign to scale. We need each campaign’s purchase revenue, refunded revenue and costs. We cannot apply the account’s 3.0x ratio to the leading campaign and pretend we have calculated its return.
Once those rows are assembled, the model tells us what to do next. Check whether the leader still leads after refunds. Compare contribution before reallocating budget. Review whether configured values reflect the product differences. Investigate the reasons behind the 38% return rate rather than assuming the ads are the whole problem.
The one-page brief I would hand over contains four things:
- The economics: a quarterly SKU-level table of revenue, refunds, costs, contribution and return rate.
- The cause: actionable refund categories, with the biggest one named.
- The assumptions: margin estimates and the costs included beside them.
- The decision: which campaigns still deserve their budgets after we account for what the business kept?
That is not another dashboard. It is a brief. The purchase event was the receipt at the beginning, not the end of the calculation. What your business kept belongs in the room where the budget gets decided.

