A purchase recorded after a ChatGPT ad interaction can later be refunded. The original advertising report remains useful evidence of attributed purchasing activity, but it may not answer how much revenue the business ultimately retained. Add a separate refund bridge before using that value to judge acquisition economics.
This guide describes an internal analysis using order and refund records. Do not assume it changes historical platform metrics or that sending a negative purchase is a supported correction method. Any platform adjustment needs its own documented procedure; the analysis can proceed without inventing one.
Define the revenue basis once
Decide what the starting purchase value includes: merchandise, tax, shipping and discounts should follow an explicit definition. Apply the same basis when subtracting refunds. If the purchase value includes shipping but the refund export only includes merchandise, the difference needs to remain visible.
Confirm the monetary scale before calculating adjustments. A refund stored in minor units cannot be subtracted directly from a major-unit revenue column. The purchase-value unit check helps prevent a representation error from being mistaken for a large return problem.
Use one analysis currency and preserve original currency values where conversion is needed. State the conversion method and date convention. An exchange-rate movement between purchase and refund may matter to finance, but it should not appear as an unexplained change in the number of refunded orders.
Keep orders and refund transactions at different grains
An order can have several refund transactions. One item may be returned today and another next week. If an order-level table is joined directly to refund rows, the original purchase value may be repeated once per refund. Aggregate refund transactions to the intended order level before combining totals.
Retain the individual refund ledger for audit, including transaction identity, order reference, amount, effective date and status. Exclude failed or reversed refunds according to the declared accounting rule. A refund request is not necessarily a completed refund, so the workflow stage must be clear.
Handle partial refunds as amounts, not as whole returned orders. A hypothetical order worth SEK 1,000 with a SEK 200 refund retains SEK 800 under the stated basis. Counting it as a fully lost order would distort both revenue and the interpretation of return reasons.
Choose the cohort view for the decision
For acquisition quality, assign refunds back to the purchase cohort they belong to. A September order refunded in October should affect the retained value of September’s acquired orders. A cash-flow report may instead show the payment leaving in October. Both views are useful, but they answer different questions.
In a hypothetical purchase cohort with SEK 10,000 of original value, known refunds comprise SEK 800 of full refunds and SEK 400 of partial refunds. The retained value at the review date is SEK 8,800. Keep the words at the review date because later returns can still change that number.
If the advertising report is available only as an aggregate, be honest about the bridge’s precision. An internal campaign reference may support a defined order cohort, but it may not reproduce the platform’s attributed population exactly. Do not subtract every business refund from a single campaign’s attributed sales and call the result a reconciled platform metric.
From order value to retained revenue
- 10,000 in purchases
Original value for the defined order cohort.
- 1,200 refunded
800 full refund and 400 partial refund.
- 8,800 retained
Internal net view as of review; further returns may remain.
Show observed adjustments separately from estimates
Known refunds and expected future returns should occupy different lines. An estimate derived from older cohorts can help planning, but it is not an observed refund transaction. State the cohort age and the evidence behind any expectation.
The return-rate sensitivity guide explores how future return assumptions change an acquisition ceiling. This page concerns the observed refund ledger and its connection to a purchase cohort. Keeping those roles separate prevents a forecast from quietly becoming a reported fact.
Review immature cohorts with care. A campaign selling products with a longer evaluation period may initially look better than a campaign whose returns arrive quickly. Compare retained revenue at similar cohort ages or display the maturity difference alongside the result.
Recalculate the right business measure
Revenue after refunds is still not contribution margin. Payment fees, return shipping, handling and unsellable inventory can change profitability beyond the refunded amount. Add those costs in the economics model using their own evidence rather than hiding them inside a revenue correction.
Preserve the original reported purchase metric beside the adjusted internal figure. OpenAI reporting documentation defines attributed purchase value and its dependence on supplied purchase events. Label your additional line as an internal retained-revenue calculation so readers do not confuse it with a native report field.
Agree a review cadence with finance and use the late-period closing process to retain versions. Each version should identify the purchase cohort, refund cutoff, definitions and unresolved linkage. The budget decision then has an auditable basis: what was originally purchased, what has actually been returned and what remains uncertain as the cohort matures.
Sources and scope
Maintain a transparent internal bridge from attributed purchase value to retained revenue after known refunds.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.