Existing non-statistical attribution models (e.g. first / last touch, linear, position, time-decay) adopt inflexible rules for allocating credit for media interactions. They do not allow for interactions between different channels or heterogenous effects (e.g. some channels may be more effective at different points in the path to purchase).
Worse, they cannot account for offline channels such as print, outdoor, and TV.
Gradient uses modern statistical causal inference methods to unpack these effects to produce useful estimates of the efficacy of each advertising channel.