Vast amounts of accumulated event data promise valuable insights for organizations through the application of process mining (PM) methods. However, extracting heterogeneous and scattered data poses great challenges. This study demonstrates a reference architecture for multimodal event log construction from fragmented event data. The artifact provides an instantiation of the reference architecture through five construction functions: 1) parsing, 2) abstraction and concretization, 3) extraction, 4) integration, and 5) enrichment, including human-in-the-loop components. Following the design science research methodology, the artifact was developed and evaluated iteratively in a real-world PM application. Customer relationship management (CRM) data, email metadata, and call transcripts were used to construct an event log of a sales process in a small use case company. Construction functions were evaluated individually as well as in terms of the overall event log constructed in collaboration with a domain expert. The evaluation demonstrates the applicability of the reference architecture to construct a multimodal event log suitable to analyze a sales process. The study contributes prescriptive design knowledge for constructing an event log with multimodal data by suggesting adaptations to the reference architecture and design objectives as well as proposing design principles.