The first ChatGPT advertising test can carry work that the following period will not need: a new landing page, an integration check and time spent learning the operating process. But campaign work does not disappear after launch. Combining everything into one cost per purchase makes it difficult to see what another period would actually require.
Start with three decisions: funding the launch, funding a comparable continuation and funding additional volume. Each needs a different view of the same cost records. The method below is an internal planning framework for advertising on ChatGPT, with hypothetical figures rather than platform prices or expected campaign results.
Define the decision before classifying the cost
A fixed cost stays unchanged within a specified time horizon and capacity range. A variable cost changes with an identified driver, such as media spending, advertisements produced or orders processed. Write that driver beside the cost. A monthly invoice is not automatically a fixed cost simply because it arrives regularly.
A service retainer might remain fixed until another campaign requires a higher service tier. That is a capacity step. Existing staff salaries might remain unchanged during a pilot while a larger programme needs overtime or outside support. Make those thresholds visible before assuming that another campaign can absorb unlimited free labour.
Cost behaviour also differs from avoidability. Commissioned creative work might be fixed and still await payment. A fixed monthly service might be cancellable before the next period. Mark each cost both by how it changes and by whether the next decision can avoid it. These classifications answer different questions and should occupy different columns.
Draw a boundary around the campaign
Include media, production, campaign-specific technology, review and operations to the extent they belong to the decision. Use invoices, agreements and realistic time estimates. Record internal time displaced from other work separately where relevant, even if no extra payment occurs. Explain that valuation instead of mixing it invisibly with supplier charges.
Product costs, payment processing and delivery also matter to profitability, but normally belong in order contribution. Do not deduct them from contribution and then add the same amounts to campaign spending. The break-even CPA calculation explains how order contribution supports a later advertising decision.
OpenAI describes the campaign budget as a spending control across its ad groups in Bidding & Budgets. Keep that control separate from the business cost map. A project authorization that includes production and operations should not be copied wholesale into the media budget.
Price the launch using four explicit rows
Consider a hypothetical test denominated in SEK, excluding recoverable VAT. Initial production and setup cost 12,000. Operations during the test cost 4,000. Assumed media spending is 18,000, and an agreed management fee equals 10% of media spending, or 1,800. Total campaign cost is therefore SEK 35,800.
Using 100 new customers solely as a planning assumption produces a fully loaded campaign cost of SEK 358 per customer. Media spending alone gives SEK 180 per customer. Both are legitimate calculations, but labelling both simply “CAC” would conceal their different boundaries. The assumed customer count is not evidence of what that ChatGPT budget will deliver.
For the next period of equal duration, assume the initial material remains usable. A fresh check costs 1,000, while operations, media and the management fee stay unchanged. Continuation costs SEK 24,800. At the same assumed customer count, that is SEK 248 each. The reduction is 11,000 rather than 12,000 because the fresh check replaces part of the original work.
Test the cost of additional volume
Now consider an addition within the available operating capacity. Assume another SEK 9,000 in media and 900 in management fees, with 50 additional customers as an explicit scenario. Incremental campaign spending is SEK 9,900, or 198 per assumed additional customer. This describes the scenario rather than the campaign’s historical average.
Check whether the expansion requires another creative version or more staffing. An extra SEK 3,000 production job would raise incremental spending to 12,900 and the scenario cost to SEK 258 per additional customer. Classification should reflect actual thresholds, rather than a preference for treating every future cost as proportional to media.
A lower average at greater volume can simply mean that setup work is spread over more customers. It does not demonstrate more efficient advertising delivery. To compare customer acquisition across channels, you also need a consistent definition of new customers. Continue with blended CAC for ChatGPT Ads once the cost boundary is settled.
One campaign, three cost questions
- First period
12,000 setup + 4,000 operations + 18,000 media + 1,800 fee = SEK 35,800.
- Next period
1,000 new checks + 4,000 operations + 18,000 media + 1,800 fee = SEK 24,800.
- Additional volume
9,000 media + 900 fee for an assumed 50 extra customers = SEK 198 each, before any capacity step.
Preserve the original cost alongside allocations
An internal model might spread the SEK 12,000 launch cost across three periods, allocating 4,000 to each. Retain the original launch amount beside that allocation. A presentation convention must not make the remaining expenditure disappear. It can help comparison without matching either the payment date or the accounting recognition period.
Document the currency conversion, tax basis and date of the estimate. Taxes that cannot be recovered remain an economic cost. Use the business’s verified treatment instead of assuming every supplier invoice has the same tax handling. Keep contingency reserves distinct from ordered work and actual expenditure, so that a reserve is not mistakenly counted twice when an invoice arrives.
Add a simple review trigger for each uncertain row. For example, revisit production once the number of required variants is known, and revisit operations when the service scope changes. An estimate without a trigger tends to survive long after its assumptions have stopped being true.
Store the map against the campaign identifier. OpenAI recommends durable IDs for integrations in Campaign Management; using them internally also prevents a renamed campaign from losing its invoice history. Assign an owner to each row so that the next budget decision starts from updated evidence rather than reconstructed memory.
Sources and scope
Identify which costs change when a ChatGPT advertising campaign launches, continues for another period or expands within and beyond existing capacity.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.