The high failure rate of demand-responsive transit (DRT) systems suggests that DRT services are viable only in selected areas. However, few studies have quantitatively examined how built environment characteristics affect DRT use. Applying gradient boosting decision trees to the data of customized bus service (CB, a type of DRT) in Dalian, we investigate the nonlinear association between the built environment and CB use, controlling for demographics and service features. Local accessibility at the residence and workplace is the most important correlates of CB use, followed by the proximity of workplace to bus stop. Some built environment variables (including distance from workplace to transit stops, distance from residence to business centers, and population density) influence CB use differently from traditional transit use observed in the literature. Furthermore, built environment variables show salient threshold associations with CB use, guiding planners to design the CB system efficiently.