Kantar's proprietary AI framework "HamiltonAI" has been trained to build models on ‘consumer response’, first and foremost, because it will always result in the best possible model.
HamiltonAI has built-in assumptions related to the model objective, that will not apply if the objective is not of a countable nature, e.g. upper bound for relationship between number of clicks and the objective count. Using revenue as the model objective, would result in the model missing information, as revenue is considered to be number of units sold X sales price per unit. Sales price per unit should however be a separate model input variable that can help explain the number of units sold.
Revenue can – in our experience – be measured in different ways and therefore be a ‘noisy’ measure as it can be impacted by many things 'out of the model's control' e.g. cost of goods sold.