Abstract
When a business sells to another business (B2B), the buying business isrepresented by a group of individuals, termed account, who collectively decidewhether to buy. The seller advertises to each individual and interacts withthem, mostly by digital means. The sales cycle is long, most often over a fewmonths. There is heterogeneity among individuals belonging to an account inseeking information and hence the seller needs to score the interest of eachindividual over a long horizon to decide which individuals must be reached andwhen. Moreover, the buy decision rests with the account and must be scored toproject the likelihood of purchase, a decision that is subject to change allthe way up to the actual decision, emblematic of group decision making. Wescore decision of the account and its individuals in a dynamic manner. Dynamicscoring allows opportunity to influence different individual members atdifferent time points over the long horizon. The dataset contains behavior logsof each individual's communication activities with the seller; but, there areno data on consultations among individuals which result in the decision. Usingneural network architecture, we propose several ways to aggregate informationfrom individual members' activities, to predict the group's collectivedecision. Multiple evaluations find strong model performance.