Measuring an AI-assisted campaign: ROAS does not tell the whole story

Article prepared with artificial intelligence assistance. The sources and method are indicated in the text.

Start with the expected decision

Should a creation be continued, a budget moved or a production method changed? These decisions require different evidence. An increase in clicks can lead to a review of a message without justifying a significant reallocation. The balance sheet should specify what the team was looking for and the scale at which it wished to decide. This preparation makes the results more interpretable and avoids choosing the most favourable figure after the event.

Understand what ROAS calculates

The ROAS reports an income allocated to an advertising expense according to the scope of the advertising. It is not automatically a margin, nor a measure of sales that would not have existed without the campaign. Two platforms can allocate credit to the same sale according to their rules. Before any consolidation, you must understand the windows, definitions and possible overlaps. An addition of dashboards does not guarantee a consistent trade balance.

What Incrementality Seeks

An approach to incrementality seeks to estimate the gap between a situation with marketing action and a scenario without this action. The relevance of the approach depends on the device, data and assumptions. A guide from the IAB published for comment on 20 October 2026 deals precisely with the joint interpretation of several methods; It is a project, not a final standard to be presented as mandatory. Our proposal is to make the issue and the limits visible in the balance sheet.

An example of calculation, not a campaign result

Suppose one entirely fictitious example: a platform allocates 30,000 money units of income to 10,000 media expenditure. The ROAS thus defined is 3. This does not mean any 20,000 profits or 30,000 additional income from advertising. Production, discounts, service costs and sales that would have taken place in any case require a separate examination. The figures are used here only to explain the calculation.

Four readings to keep separate

Measure time to a usable version

Generation time is a production step, and the full cost must include preparation of the brief, selection, corrections and validations. An agency can quickly produce many proposals, few of which will be usable. The pilot must therefore increase the number of versions selected, the causes of rejection and the time to delivery. Comparison with the usual method should be based on an equivalent quality requirement and deliverable.

Keep the creative test context

A comparison becomes difficult when the offer, targeting, budget and message change simultaneously. The team must document these developments and avoid a conclusion too precise if the device does not allow their role to be isolated. A new creation can also contribute to brand recognition on a different horizon from immediate conversion. The briefing should provide for this dimension rather than asking for a single figure to judge all the objectives.

Write a conclusion that can be discussed

A useful balance sheet gives the observed result, its scope, its limits and the proposed decision. It also indicates the information that is missing to go further. Artificial intelligence can help to reconcile files or prepare a synthesis; totals, reconciliations and conclusions must be checked. An assured formulation does not replace a protocol. The role of the agency is to explain what the data allow for decision and where judgment should be kept prudent.

Sources and method

• IAB, Cross-channel Measurement Decision Guide, draft submitted for comment on 20 October 2026
• NIST, Generative AI Profile AI 600-1

The recommendations and grids of this article are proposed by Le Fils de Pub. The constructed examples are indicated; the advertisements and results attributed to an organization remain those of this source. Article prepared with artificial intelligence assistance.