Poor coordination in human-robot collaboration can lead to inefficiencies and, more critically, to risky situations for human operators. Such coordination issues often stem from task planning that overlooks the presence of humans, who impact the duration of the robot's actions due to safety measures, e.g., if they have to access the same area simultaneously. This paper proposes an approach that leverages information from past process executions to estimate the coupling effect between actions performed concurrently by humans and robots. We introduce a synergy coefficient for each human-robot task that quantifies how human actions affect the duration of robotic actions. We implement the proposed method in a simulated scenario where agents share a collaborative workspace. We show that our approach can learn such bad couplings, enabling the enhancement of a task planner with this information, fostering a proactive agent interaction.