A ChatGPT campaign report can show a cost of SEK 250 per purchase while the advertiser’s test costs SEK 400 per purchase. Both numbers can be correct. Agency setup, ongoing work and contractual percentage fees may explain the difference. Confusion starts when those numbers share a label or the allocation method changes after the outcome is known.
Decide how fees will be treated before the test starts. The purpose is to compare media efficiency with the advertiser’s full investment in this specific test. The method below is an internal management recommendation, not an OpenAI billing rule or a statement about typical agency prices.
Keep two measures with explicit names
Use media CPA for advertising spend divided by the conversions under evaluation. Use total test cost per purchase for media plus agency charges assigned to the test under your chosen method. Avoid calling the second measure total CAC if the denominator includes returning customers or other acquisition expenses have been omitted.
OpenAI Reporting describes advertising spend and CPA. An external agency invoice needs a separate calculation in the advertiser’s own records. Preserve the original platform measure and add clearly named financial columns alongside it, so that someone reviewing the result later can identify the cost boundary.
Use the same campaign period, currency and outcome definition for both measures. If the test is assessed using qualified leads, divide the full cost by qualified leads as well. Comparing purchases in one table with all form submissions in another creates a difference that fee allocation cannot explain.
Assign work before assigning percentages
Start with costs directly attributable to the ChatGPT test. These might include a separately commissioned setup or work on a destination used only for that test. Direct costs normally do not require an allocation key. Still record whether the deliverable will support other campaigns later, because that matters when assessing future periods.
Shared work needs a defensible allocation method. Recorded hours, agreed deliverables or an advance agreement about resource share can all provide a basis. Choose the measure that best describes the work consumed. Allocation by media spend is convenient, but a small new test may require much more setup than a large established campaign.
Across the relevant assignments, allocated amounts should reconcile to the shared invoice. A charge assigned to ChatGPT must not remain fully assigned to another channel too. Explain which work is outside the test and why, so a low total does not depend on invisible support from another budget.
A worked example with three fee components
Assume a hypothetical test with SEK 10,000 in media spend and 40 purchases under a specified measurement definition. All costs are in Swedish kronor excluding VAT. Direct setup costs SEK 2,000. A shared monthly retainer is SEK 12,000, and the documented work supports assigning 25% to this test. That contributes SEK 3,000.
The contract also charges 10% of media spend. In this example the fee base is exactly SEK 10,000, creating a variable charge of SEK 1,000. Total agency charges assigned to the test are SEK 6,000. Adding media produces SEK 16,000, or SEK 400 per purchase. Media CPA remains SEK 250.
These invented figures demonstrate the calculation; they are neither market rates nor campaign results. The SEK 400 figure also needs a suitable economic comparison. Product gross margin alone may omit fulfilment or other order costs. Use a consistent contribution definition, as described in the economic CPA ceiling guide.
Four lines that make agency costs visible
- Direct setup
SEK 2,000 belongs entirely to the test. Identify it as one-time work.
- Shared retainer
25% of SEK 12,000 gives SEK 3,000, supported by a documented share of work.
- Variable fee
10% of SEK 10,000 media spend gives SEK 1,000. The fee base is media alone.
- Total test cost
10,000 + 2,000 + 3,000 + 1,000 = SEK 16,000. With 40 defined purchases, cost is SEK 400 per purchase.
Read the percentage-fee base literally
Ten percent is incomplete without a base. A contract might specify media spend, spend after credits, or another explicit amount. Check how it treats discounts, currency conversion, minimum charges and any fee tiers. Do not assume such conditions exist; record them where they actually apply.
If the fee is 10% of media, do not calculate 10% of media plus agency charges. That produces a different fee from the one specified. A minimum charge can also make the effective rate higher on a small test. Show both the contractual rate and the amount that actually belongs to the test.
Working backwards from the total allowance provides a useful check. Suppose, hypothetically, the test may cost SEK 18,000, allocated fixed fees are SEK 5,000 and the variable fee is 10% of media. The amount available for media is 13,000 divided by 1.10, approximately SEK 11,818.18. Spending the entire SEK 13,000 remainder on media would leave no room for its percentage fee.
Separate the completed test from the next decision
The completed test should show its setup cost. The next-period decision should ask which costs genuinely follow from continuing. An already paid setup may not affect that decision, while ongoing analysis, production and a minimum retainer may still be necessary. Identify the specific work that changes instead of assuming every agency cost disappears after launch.
Present two columns: the full completed test and the proposed continuation. Preserve the first column. Removing setup retrospectively to make the initial test look more attractive is different from preparing a forward-looking estimate. The guide to fixed and variable campaign costs helps define these assumptions.
Also consider reusable work carefully. A measurement plan created during the test may benefit later campaigns, but allocating it across imagined future activity can understate the current commitment. If finance uses a formal allocation period, show that policy and the original cash cost separately. Do not invent a longer useful life simply to pass a performance threshold.
Leave a trail that a reviewer can follow
Keep invoice period, cost category, allocation key, percentage base, currency and owner with the report. Treat credits and late invoices consistently between periods. Mark the total as provisional if the agency charge is not final, and retain the earlier estimate when it is updated.
The agency and advertiser can then discuss two concrete questions: whether the media buying performs efficiently, and whether the complete operating approach is economically worthwhile. Both questions matter. Clear cost ownership lets the conversation proceed without changing what CPA means from one meeting to the next.
Sources and scope
Choose and document a defensible allocation of agency charges when evaluating a specific ChatGPT advertising test.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.