Current soccer evaluation models such as VAEP or Expected Threat (xT) have advanced performance analysis by quantifying the impact of player actions on future scoring probabilities. However, these models rely on outcome-driven learning, which conflates execution quality with tactical intent. As a result, successful but risky actions are often overrated, while strategically sound actions that fail are undervalued—a distortion known as outcome bias. In this paper, we introduce xSuccess, a probabilistic model that estimates the likelihood of successful action completion based on contextual features. We integrate this completion probability into an adjusted action value framework, exemplified using VAEP. Our method isolates completion risk from estimated scoring impact and produces \({VAEP_{adjusted}}\) values that reflect both reward and uncertainty. Using a public dataset of over 3.6 million on-ball actions, we demonstrate that xSuccess is well-calibrated and the adjusted values align with Expected goals (xG) distributions. Comparative analysis shows that the adjusted model reduces variance caused by rare successful events and offers more realistic valuations. A case study of a goal sequence from Borussia Dortmund illustrates the practical relevance of this framework. Our findings suggest that completion probability is a vital component in soccer analytics, enabling a more accurate and interpretable assessment of player performance under real-game conditions.