Value-based decisions are an integral part of our life. Many disorders are characterized by aberrant choices with, at times, detrimental consequences for the physical and mental health of an individual. In the past decades, the field has progressed by optimizing tasks and computational models to better describe canonical decision-making processes. However, concerning clinical applications, important gaps remain that relate to various limitations of conventional laboratory-based assessments. In this chapter, we will outline opportunities and challenges of leaving the lab to attain more naturalistic applications that may ultimately help improve the prediction of behavior and the classification of disorder-related states. To this end, we will discuss psychometric principles, provide examples for suitable designs, and briefly review hierarchical modeling approaches to exploit the information contained in large, naturalistic data sets. Consequently, these emerging lines of research offer great promise for computational psychiatry in monitoring treatment effects over time and tailoring therapy-related decisions to changes in latent decision-making processes.