UserJourney2Vector: Enterprise Application of Transformer Models on User Telemetry Data

User telemetry data offers valuable insights into user behavior and goals when interacting with digital products and services. However, analyzing vast amounts of event sequence data can be challenging. In this case study, we apply transformer neural networks to telemetry event sequence data, yielding the UserJourney2Vector model, which treats event sequences similarly to natural language to distill complex user journeys into latent vector representations. The model enables understanding of typical and anomalous user paths, prediction of next actions, and user segmentation based on behavior. We used the model to obtain user clusters, predictive journeys, and cluster statistics, then conducted interviews with digital product experts to assess potential applications and impact of the model output. Experts responded positively to the model's ability to bolster user personas with data-driven insights and noted integrating model outputs with current practices could augment product design and data interpretation.