In some embodiments, model 300 may predict a goal or intent of a user. This goal or intent may be selected from a plurality of goals and/or intents stored by the system. Model 300 may first determine an intent cluster (e.g., a group or category of intents) and then select a specific intent from the intent cluster. In some embodiments, the system may determine the cluster of intents based on the similar feature inputs. For example, the system may cluster goals/intents based on similar characteristics of the users. For example, the system may determine that users who ask different questions about payment have similar account information and digital activities. The system may further determine that the users tend to be different from those of users who have a one-off type request, such as lost card reports or travel notification.

A multi-tiered approach may be used to capture this behavior. The first layer of the model (e.g., model 320) identifies which group of goals is most likely, then in the subsequent layer, the model (e.g., model 330) identifies which specific goals are most likely. The clusters of goals used in the first layer (e.g., model 320) are derived based on feature data and the known goal/intent list, which can change as available data changes or expands. In some embodiments, a specific intent may comprise its own intent cluster and/or not every potential specific intent needs to belong to an intent cluster. For example, if the first-layer model (e.g., model 320) determines that none of the existing clusters are likely, a default classification model may be used to make a prediction at goal level to make sure that goals not belonging to any cluster can be predicted.