Individuals with motor-impairments struggle to consistently perform touchscreen gestures, resulting in unsuccessful or unintentional interactions with devices. Existing touch models rely on static definitions of a user, thus fail to capture the variable and erratic nature of their interaction abilities. We conducted a four-week in-the-wild user study to measure the touchscreen performance abilities of nine motor-impaired people. Our results identified significant variance in touchscreen performance between interaction sessions. We proposed novel techniques to measure and model individuals’ abilities from otherwise discarded interaction data. Using these measurements we were able to create session-specific models, from other users’ data, that significantly outperformed existing user-specific models.