Google patents ad system that adjusts bids for new customers

1. A method, comprising: receiving, by one or more processors, digital component campaign information comprising an initial conversion value, one or more additional conversion values, an initial bid value, one or more additional bid values, and ground truth data identifying one or more qualifying consumers; receiving, by the one or more processors, a request for a digital component; generating, by executing a first stage model of an artificial intelligence (AI) model trained using the ground truth data, one or more consumer embedding representations; generating, by executing a second stage model of the AI model trained using the ground truth data, one or more merchant embedding representations, wherein at least one of the first or second stage model is trained offline; providing, by the one or more processors, as input to the AI model the request for the digital component, the one or more consumer embedding representations, and the one or more merchant embedding representations; determining, by the one or more processors executing the AI model online, at least one of a new consumer probability or a new qualifying consumer probability, wherein: the new consumer probability corresponds to a probability a consumer to be presented the digital component is a new consumer, and the new qualifying consumer probability corresponds to a probability the consumer to be presented the digital component is a new qualifying consumer; and dynamically adjusting, by the one or more processors based on at least one of the one or more additional bid values or the one or more additional conversion values, and at least one of the new consumer probability or the new qualifying consumer probability, at least one of a resulting bid value or a resulting conversion value associated with the digital component.