Compared to single-turn or multi-turn dialogue, persuasive dialogue is more challenging as it aims to influence users toward a specific goal over a dialogue session. Previous works mainly focus on dialogue-related elements such as relevance and persuasive strategies while neglecting human factors. In this paper, we integrate human factors into both the input and output sides of the model to guarantee that users can be guided smoothly toward the persuasion goal. On the input side, we model the will scores of users towards the persuasion goal. Inputting these signals into the persuasion model enables an understanding of persuasion difficulty levels and personalized persuasion strategy selection. On the output side, we use real-world user persuasion outcomes to guide model training by formulating the problem using reinforcement learning. Both online and offline scenarios are considered in our formulation. To facilitate human-factor-related research, we collect a dataset that includes real user profiles, dialogues, and persuasion outcomes from real-world applications. Experiments show that our method achieves an improvement of 11.6% over the baseline in terms of average ratings in human evaluation.