Lectura de tesis
Jueves, 19 de Noviembre de 2026 10:30
- Título de la tesis: Supporting End Users in Knowledge-Intensive Digital Tasks: Design Guidance for Board-Based Tools and Behavioral Insight from Event Data
- Doctorando: Alfonso Bravo Llanos
- Directores: Manuel Resinas Arias de Reyna y Joaquín Peña Siles
- Tutor: Joaquín Peña Siles
- Resumen: Flexible work management tools such as Trello, Asana, and Notion are generified by design: rather than embedding a domain model, they provide generic capabilities that users adapt through structures, workflows, views, and conventions. Work once done by analysts and consultants has shifted to end users, who configure these tools through the same interfaces in which work is performed, while the event data these tools record offers them further opportunities to understand and assess how work proceeds. Configuring these tools and making sense of the data they record are knowledge-intensive digital tasks that call for design and analytical expertise that end users do not necessarily possess, which limits how effectively they can perform them. Existing research offers relevant foundations in generification, end-user development, process mining, and process performance measurement, but leaves these tasks without direct support.
This thesis investigates how end users can be supported in such tasks, while preserving the contextual knowledge and judgment only they can bring. The first part addresses the design of boards in Board-Based Tools (BBTs), a class of flexible work management tools. It systematizes the BBT design space through a metamodel and eight design patterns, then examines how this knowledge can guide users toward an effective board design, using Large Language Models (LLMs) to support its application. Results show that explicit design knowledge improves design quality, while LLM-based guidance achieves comparable quality in less time.
The second part turns to knowledge derived from digital traces. Analyzing board event logs reveals how boards are actually used and how they have evolved, giving users evidence to assess redesigns they could otherwise judge only from memory. Beyond BBTs, LLMs generate candidate Process Performance Indicators (PPIs) grounded in observed behavior from process event logs and translate them into definitions that can be computed automatically over event data, leaving to users the judgment of which candidates matter. Results indicate that behaviorally grounded PPIs are more relevant and less redundant than those generated from a process description alone.
Overall, these contributions show how explicit design knowledge, automated analysis, and LLM-based guidance and suggestions can compensate for the design and analytical expertise these tasks demand, without replacing the decisions or the knowledge of the work that remain the end user’s.