Features

Panel MMM

Hierarchical partial-pooling across time series, single panels like geographies, or multi-dimensional panels like geography and brand

Adstock and saturation

Geometric, Weibull, logistic, and Michaelis-Menten transforms.

Time-varying parameters

HSGP-based (Hilbert Space Gaussian Process) intercepts and media effects to capture structural breaks.

Mundlak / correlated random effects

Within-unit mean corrections to address unobserved panel confounding.

Budget optimization

Constrained non-linear allocation with response sampling.

Pipeline runner

Staged artifact generation with machine-readable manifests for reproducible deployments.