An order that looks profitable when the purchase is recorded can produce negative contribution when the product comes back. For advertisements on ChatGPT, the evaluation date matters. The advertising cost has already been incurred, while the refund, return shipment and inspection of the product may arrive later.
A useful sensitivity analysis shows how alternative return rates change contribution per original order. It does not assume a universal return rate for ChatGPT customers. Use the business’s own mature order cohorts where available, and label other rates as scenarios. The method below is an internal planning exercise with hypothetical amounts.
Define the return rate you are testing
Start with a bounded group of original orders, such as purchases placed during a campaign period. In the simple model below, each order contains one product that is either kept or fully returned. The return rate is fully returned orders divided by all original orders. Refunded revenue as a percentage of sales is a different measure and cannot simply replace it.
Keep cohort membership fixed as returns arrive. Removing returned orders from the denominator while distributing their expenses across the remaining orders makes the result harder to compare with cost per recorded purchase. You can also report cost per kept order, but give it a separate label and denominator.
Real baskets containing several items often need more detail. Partial returns, exchanges and different product margins do not always fit a binary model. Preserve order and item identifiers so that a later refund can be connected to its original purchase rather than assigned to whatever campaign happens to be running that week.
Write two complete order outcomes
Assume a hypothetical product sells for SEK 1,000 excluding VAT. Product cost is 400, outbound delivery and picking cost 70, and the payment fee is 30. If the customer keeps the product, contribution before advertising is 1,000 minus 400 minus 70 minus 30, or SEK 500.
For a full return, all product revenue is refunded. Assume the product is restored to saleable inventory with its full SEK 400 cost value intact. The product’s net cost against the completed order is therefore zero. Outbound delivery and the payment fee are not recovered. Return transport and handling add another SEK 80. The returned order contributes negative SEK 180.
That loss consists of 70 plus 30 plus 80. Do not subtract the full selling price again from negative 180; the refund has already reduced net revenue to zero. Likewise, do not charge the product cost again after recovering its full inventory value. Those two double counts can turn a reasonable return scenario into a misleadingly severe loss.
Compare three possible outcomes
At return rate r, average contribution before media is 500 × (1 − r) − 180 × r. This simplifies to 500 − 680 × r. Each additional percentage point of returns therefore reduces contribution by SEK 6.80 per original order under these specific assumptions.
At 10% returns, contribution is SEK 432. At 25%, it is 330; at 40%, it is 228. With hypothetical media CPA of SEK 300, the remaining amounts are respectively 132, 30 and negative 72 per original order. None of these rates or outcomes is a claimed campaign result.
The mathematical break-even rate after media is 200 divided by 680, approximately 29.4%. However, breaking even after media leaves nothing for shared fixed costs. If the business requires at least SEK 60 remaining, the corresponding threshold is 140 divided by 680, approximately 20.6%. Use the break-even CPA guide to connect that contribution requirement to the wider acquisition decision.
Returns change the economics of the same purchase
- 10% returned
0.90 × 500 − 0.10 × 180 = SEK 432. After media, SEK 132 remains per original order.
- 25% returned
0.75 × 500 − 0.25 × 180 = SEK 330. After media, SEK 30 remains per original order.
- 40% returned
0.60 × 500 − 0.40 × 180 = SEK 228. After media, the shortfall is SEK 72 per original order.
Check whether inventory value really returns
Full inventory recovery is a consequential assumption. If a returned product in the same example loses SEK 100 of value, the returned order contributes negative 280. At a 25% return rate, average contribution falls from 330 to 305. Only SEK 5 remains after the assumed media cost.
Use realistic recovery value and handling costs for the product group. A product that can be resold should not receive both its full original cost as a loss and an additional write-down for the same damage. If it has no remaining value, its entire product cost instead belongs in the returned-order loss. Record each economic loss once.
Check customer shipping payments and which amounts were refunded as well. The simplified example assumes no separate shipping revenue. The shipping subsidy calculation covers other arrangements. Keep all amounts on a consistent tax basis; a customer payment including VAT cannot be compared directly with costs excluding recoverable VAT.
Follow the cohort until returns mature
A cohort delivered two days ago has not had the same opportunity to return products as an older cohort. Compare at the same age after delivery, and inspect how long returns keep arriving in the business’s records. That pattern can differ by size, product and offer.
Show recorded returns alongside a separate estimate for unresolved orders. As actual returns become known, they replace the estimate rather than being added on top of it. Retain previous forecast versions so that future campaign planning can learn from systematic overestimation or underestimation.
OpenAI’s Reporting describes attributed conversion measures. Those do not replace an order and return ledger. Verify the purchase-event definition against Conversion Tracking, and do not assume that an economic refund adjustment automatically appears in the report being used.
The resulting decision should identify the product group at risk of crossing the internal boundary and the date when the cohort will support another review. If repeat purchases are expected to offset return costs, support that claim with observed customer contribution over time. Otherwise the model merely moves an uncertain assumption from the returns column into customer value.
Sources and scope
Identify the return rate at which a ChatGPT campaign’s order contribution no longer covers media spending and the business’s required remaining margin.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.