Several platforms claimed the same revenue.
The retail brand spent around €90,000 a month on media, and several platforms claimed profitable revenue from overlapping customers. Adding their reported contribution produced more revenue than the business had actually earned. Checking the records alone could not answer where the next budget increase should go.
Acquisition cost and reporting effort
During the twelve-week controlled budget test, acquisition cost fell 14% while new-customer volume remained broadly comparable. Reporting time fell 60%, and allocation recommendations documented their evidence and uncertainty.
The 12-week budget test describes a 14% reduction in acquisition cost with broadly comparable new-customer volume. The reporting saving is a separate process improvement. The case supports a more disciplined allocation framework, rather than a claim that one attribution view became universally correct.
Give each source of evidence a clear job
Customer and order records showed the business outcome. Platform reports helped with campaign optimization. A limited allocation test examined whether the next budget change created additional value.
The review kept the comparison, timing and confidence visible. Where the evidence supported only a direction, the recommendation said so and proposed the next check before a larger commitment.
Give the next budget change a testable question.
Platform reports could each claim the same sale. A limited allocation test used customer records, comparable volumes and a review window long enough for purchases to appear.
| Evidence | Useful for | Insufficient on its own for |
|---|---|---|
| Platform credit | Optimizing delivery within a platform’s reporting rules | Claiming every credited customer was incremental |
| Order and customer economics | Checking actual volume, cost and customer mix | Isolating one channel’s causal contribution |
| A limited comparison | Evaluating the consequence of a defined budget change | Assuming the result transfers to every spend level |
| Agreement across imperfect signals | Choosing a cautious action when a clean test is infeasible | Presenting triangulation as experimental proof |
The decision concerned the next investment. Attractive historical channel averages could not answer it alone.
Reading the controlled budget test
Experiments were constrained by audience size, spillover, timing and platform learning. Where a clean causal comparison was not feasible, the recommendation needed to reflect triangulated evidence and its uncertainty instead of promoting observational data to proof.
Retargeting and demand harvesting still had operational value, but their attractive attributed returns were not sufficient grounds for unrestricted expansion. The investment decision needed to consider what changed when spend changed.
Attribution remained an estimate. A limited spending change gave the business evidence about its next investment without requiring a perfect account of every customer touchpoint.