Reallocate ad spend with causal response curves

To place the next unit of advertising spend, a budget owner needs marginal return rather than average return. Average ROAS can overstate channel value when demand raises both spend and revenue or when a mature channel has reached saturation.

The case estimates a response curve for each channel and market. It then compares marginal return at the current spend level across the full portfolio.

To separate advertising effect from demand, the method models the demand factor that influences both budget and revenue. Each allocation carries an uncertainty interval and a bootstrap vote share, which make model agreement and sensitivity explicit.

To test accuracy against a known answer, the validation uses semi-synthetic data with a defined dose-response curve. The comparison includes a naive method, a generalised propensity score method, a temporal deconfounder, and the eomer method.

The fixed-budget allocation estimates 13% more attributed revenue, with a 90% interval from 12% to 29%. In the validation, the eomer method records nine to 53 times lower error than the naive method across the two reported regimes.

The allocation remains a model estimate. Large budget changes should proceed through a controlled geographic or audience holdout test before the team applies the full reallocation.