Two channel managers arrive at the budget meeting with good news. Search reports $12 in credited sales for every advertising dollar. Social reports $5. If the question is which screenshot looks stronger, the meeting is over.
If the question is where to put the next $2,000, the work has barely begun.
Advertising credit, unique customer orders, retained contribution and cash available to spend describe different things. Treating them as interchangeable can reward the channel that reports the most attractive number while concealing the customer problem that makes those sales expensive.
Put the two channels on a declared comparison basis
Illustrative example. All businesses, customers, channel reports, costs and outcomes below are fictional teaching assumptions, not Aright results or industry benchmarks. The deliberately different refund rates make the economic effect visible; they are not typical rates for search or social advertising.
Consider one store selling one product in one market, in US dollars. Two campaign groups run over the same acquisition period. There are 400 paid, delivered orders, each for one $100 unit. For this constructed example, there are no discounts, cancellations, undelivered orders, exchanges or partial refunds, and all campaign media costs in scope are included.
Every order has completed the same 45-day observation period after delivery. This fixes the age of the data, not anyone's return or legal rights. The figures describe the chosen cutoff; later claims or corrections can still require a restatement.
The team groups 200 orders under A, paid search, and 200 under B, paid social, using one declared last-paid-click rule in its operational records. Every order appears once in this planning view. That is a way to organize the comparison, not evidence that the last click caused the purchase.
The original advertising views remain separate. In the fictional extracts, A credits itself with $24,000 of sales and B with $20,000. A also credits 40 orders, worth $4,000, that the exclusive planning rule puts in B. The same sales therefore appear in both native views.
| View or measure | A: paid search | B: paid social |
|---|---|---|
| Sales value credited in each native advertising view | $24,000 | $20,000 |
| Media spend | $2,000 | $4,000 |
| Native credited-value/media ratio | 12× | 5× |
| Unique gross order value under the operational grouping | $20,000 | $20,000 |
| Gross order value/media on that common basis | 10× | 5× |
The business has $40,000 in unique gross orders, not $44,000: $24,000 + $20,000 − $4,000 of overlap. It has not discovered that a platform's attribution model is “wrong.” It has discovered that two credit views cannot simply be added to count company sales.
Google Analytics describes attribution as allocating credit among interactions on a path. Keep the model, window and reporting scope with the original numbers. The exclusive planning view answers a different question and should not silently replace those definitions.
Likewise, Google Ads' Target ROAS explanation relates reported conversion value to advertising cost. What is sent as conversion value matters. A gross basket amount is not automatically revenue retained after refunds, and neither is a complete contribution calculation.
Follow the same orders through refunds and costs
At the cutoff, A has 80 fully refunded orders out of 200, or 40%. B has 10 out of 200, or 5%. All 90 returned units have been physically received, and all refunds in this example are completed. There are no open cases in these two groups.
For each delivered unit, assume a $40 goods cost, $5 outbound delivery cost and $3 non-refundable payment processing cost. Every returned unit adds $8 return freight and $2 inspection. Its assumed net recoverable value is $30 after future reconditioning and resale costs, excluding the separately listed inspection charge. This is an economic credit, not cash already available or another full-price sale.
The following internal model excludes tax and duty effects, fixed overhead and other unlisted costs. Real market routes need those omissions checked. It is not an accounting profit statement or a platform metric.
| Cohort calculation | A: paid search | B: paid social |
|---|---|---|
| Gross order value | $20,000 | $20,000 |
| Full refunds reversing sales revenue | −$8,000 | −$1,000 |
| Sales revenue retained at cutoff | $12,000 | $19,000 |
| Goods cost of all 200 delivered units | −$8,000 | −$8,000 |
| Original outbound delivery | −$1,000 | −$1,000 |
| Non-refundable payment processing | −$600 | −$600 |
| Return freight | −$640 | −$80 |
| Inspection | −$160 | −$20 |
| Net recovery credit | +$2,400 | +$300 |
| Contribution before media | $4,000 | $9,600 |
| Media spend | −$2,000 | −$4,000 |
| Contribution after media | $2,000 | $5,600 |
For A: $12,000 − $8,000 − $1,000 − $600 − $640 − $160 + $2,400 = $4,000 before media, then $2,000 after media.
For B: $19,000 − $8,000 − $1,000 − $600 − $80 − $20 + $300 = $9,600 before media, then $5,600 after media.
A second check makes the model easier to inspect. A retained order contributes $100 − $40 − $5 − $3 = $52 before media. A full return contributes $0 − $40 − $5 − $3 − $8 − $2 + $30 = −$28. A therefore has 120 retained orders × $52 plus 80 returned orders × −$28: $4,000. B has 190 × $52 plus 10 × −$28: $9,600.
The refund has already reduced sales revenue. Do not deduct it a second time as another expense. The goods cost includes all delivered units, so do not also write off the same returned cost without reconciling the recovery treatment.
Google Analytics' metric definitions distinguish gross purchase revenue, purchase revenue after refunds and refund amounts. Even when a field already subtracts refunds, it does not supply this store's complete cost model. Reading the definition prevents both leaving refunds out and taking them off twice.
The historical interpretation changes—but it is still historical
A has the higher common-basis gross sales/media ratio: 10× against 5×. B has the higher contribution after media: $5,600 against $2,000, despite twice the media spend.
It is useful to compare efficiency as well as totals. Historical contribution after media per media dollar is $2,000 ÷ $2,000 = 1.0 for A and $5,600 ÷ $4,000 = 1.4 for B. These are explicitly defined internal ratios, not ROAS. They explain why, in this cohort, A's attractive gross-sales number does not support giving it more budget automatically.
The next operational question is why A has so many refunds. The same product and price do not mean the same customer expectation. A promotion might attract the wrong use case, a promise could be unclear, or an operational problem might be concentrated in the relevant orders. Those are hypotheses to investigate, not properties of search advertising.
If A's refund problem can be corrected, its economics may change substantially. With all other illustrative inputs fixed, 40 returns instead of 80 would produce 160 × $52 − 40 × $28 − $2,000 = $5,200 after media. That is still less total contribution than B's $5,600, but $2.60 of contribution after media per advertising dollar, compared with B's $1.40. The preferred historical comparison depends on whether the constraint is total spend, fulfilment capacity or total contribution. The scorecard should expose that choice rather than hide it in a winner badge.
Contribution is not the money available for tomorrow's campaign
Suppose B's payment provider has a $2,000 reserve that is not yet released. A higher contribution estimate does not make that money spendable. Equally, withholding a reserve does not by itself create another $2,000 operating expense. Keep the economic view and the cash-timing view separate.
In this fictional planning meeting, the business has $2,000 of unrestricted cash after protecting existing customer-remedy commitments, but only 20 additional units of fulfilment headroom. There is a proposal to reserve up to $500 for a B test, capped at 20 orders.
The base prefunding alone could be $800 for goods, $100 outbound delivery, $60 payment processing and $500 media: $1,460 before counting new customer receipts. The remaining $540 is not proof that the test can fund refunds. New orders may create new refund and return costs while their payments are held; a reserve for existing customers is not automatically available for those new liabilities. The $30 recovery credit per returned unit is also not cash in the bank.
Finance must confirm settlement timing, holds and funding for the new cohort's remedies before the spend is released. If that cannot be funded, the decision is to delay or reduce the test, even though B's historical contribution looks better. Do not spend the same reserve once in the scorecard and again in the campaign plan.
The next-dollar question needs its own evidence
Even after reconciliation, there is a difference between “B had a better cohort” and “moving money to B will increase company contribution.”
A search campaign could be serving people already persuaded elsewhere. Social might be creating demand that later appears under search. B might also have exhausted its easiest audience: the next $500 could cost more per retained order than the earlier $4,000. A last-click grouping cannot settle those questions.
Google's Conversion Lift documentation describes controlled exposure comparisons for assessing incremental outcomes and notes that the tool is not available to every account. That is a different evidence task from reading an attributed-sales report. It does not establish that this store has access, enough observations or a suitable experiment.
For the example business, the budget decision should be written as follows:
- Do not increase A merely because its gross-sales ratio is highest. Diagnose the refund mechanism and correct misleading promises or fulfilment faults promptly.
- Earmark a maximum $500 B test and 20-order operating cap, conditional on funding new customer remedies and enforcing the service limits. Neither number is a predicted result.
- Define the comparison before spending. Where feasible, use a suitable holdout or controlled allocation to ask whether the change increases whole-business retained contribution, including interactions with A. Do not judge it solely by the platform's own credited conversions.
- If a credible comparison is unavailable, call the exercise a capped operational/measurement pilot. It may reveal whether tracking, cost capture and fulfilment work; a before-and-after improvement alone will not justify a causal scaling claim.
Judge the new cohort only after its stated observation period, while monitoring service problems immediately. Stop or change the pilot when cash backing, stock, dispatch or remedy capacity fails—not after another attractive sales screenshot arrives. Late refunds and cost corrections should update the version of the scorecard used for the decision.
Keep the scorecard small enough to explain
At the end of the meeting, management needs to understand five linked questions: what each channel claimed, what the store actually sold once, what remained after refunds and stated costs, what cash and operating capacity are available, and what evidence is needed for the next allocation.
This example uses one store to make the reasoning visible. A marketplace settlement would introduce its own deductions and timing; another market could introduce currency, duty, tax or fulfilment differences. Reconcile those branches explicitly rather than treating the amount paid into an account as gross sales or mixing incompatible routes into a regional average.
The useful result is not a permanent verdict on search versus social. It is a decision that survives the obvious objections: preserve the source views, remove duplicate counting from the company sales view, confront the refund economics, protect customer service, and test the next investment at a size the business can support. A scorecard earns its place when it changes that decision—not when every dashboard number has been copied into one table.