Channel efficiencies — the truth the model must find
Set the simulated channel strengths. Compare the model’s estimated contributions with the known simulated truth.
The world
samples:
Every change simulates 208 weeks and runs River-1 on the observed history. Dashed lines are the simulated ground truth; shaded bands are 80% posterior intervals.
Explore a synthetic marketing world. River-1 decomposes simulated sales into baseline, channel and control contributions. Adjust the true channel strengths and compare the posterior with the known truth. Results can be wrong: this is one artificial example, not validation on real marketing data.
—
highest contribution / spend
—
mean ratio error vs truth
0
worlds fitted this session