This disclosure relates to dynamic determination of data-transmission characteristics based on historical data accesses. For example, some embodiments include accessing a content stream having (i) a first content item and a second content item to be provided to one or more user devices associated with a user and (ii) a trigger condition for providing the second content item. In addition, some embodiments include detecting that the first content item has been consumed at the one or more user devices through a first communication channel and, further, determining that the trigger condition has been met. A machine-learning model inputs action data to determine a second communication channel from among two or more available communication channels. In some cases, the second content item is triggered to be delivered via the second communication channel, responsive to the first content item being consumed and the trigger condition being met.