Abstract: Techniques described herein include systems and methods for throttling requests for content to reduce stress on a check out pipeline associated with an electronic marketplace thereby avoiding overstressing a server to the point of no longer processing requests from users. In embodiments, first information may be maintained that identifies a ranking for a plurality of items based on a score. A predicted velocity of content requests about the plurality of items may be maintained and second information about an actual velocity of content requests about the plurality of items may be received. In response to a request for content, a portion of items may be identified based on the scores associated with said portion and partition the portion into a number of groups or partitions based on the predicted velocity and the second information. A data object that comprises the portion of items associated with a partition may be generated.

Abstract: Techniques for providing a messaging plan may be provided. A message that conforms to a channel agnostic format may be received. In an example, a messaging plan that includes a channel for delivering the message may be determined based at least in part on a score determined by a machine learning module utilizing metrics of a user. The message may be provided to the user according to the messaging plan. The messaging plan may be updated for subsequent messages based at least in part on updated metrics of the user received in response to the user interacting with the message.