In parallel order-taking practice, tasks should be batched by actual geographic location rather than selecting tasks by platform, in order to plan efficient routes and complete tasks from multiple agents in one trip.