Software startups develop innovative, software-intensive products or services. Such innovativeness translates into uncertainty regarding a matching need for a product from potential customers, representing a possible determinant reason for startup failure. Research has shown that experimentation, an approach based on the use experiments to guide several aspects of software development, could improve these companies' success rate by fostering the evaluation of assumptions about customers' needs before developing a full-fledged product. Nevertheless, software startups are not using experimentation as expected. In this study, we investigated the reasons behind such a mismatch between theory and practice. To achieve it, we performed a qualitative survey study of 106 failed software startups. We built the eXperimentation Progression model (XPro), demonstrating that the effective adoption and implementation of experimentation is a staged process. First, teams should be aware of experimentation, then they need to develop and intention to experiment, perform the experiments, analyze the results, and finally act based on the obtained learning. Based on the XPro model, we further identified 25 inhibitors that prevent a team from progressing along the stages properly. Our findings inform researchers of how to develop practices and techniques to improve experimentation adoption in software startups. Practitioners could learn various factors that could lead to their startup failure so they could take action to avoid them.