As a data engineer responsible for managing ETL (Extract, Transform, Load) pipelines, you have observed that a streaming pipeline running on Google Cloud Dataflow is experiencing significant delays in processing incoming data, leading to output delays. Upon inspection, you found that Dataflow automatically optimized the pipeline graph, merging various steps into a single step. To resolve the issue, you need to identify the potential bottleneck in the pipeline. What steps should you take to diagnose the source of the delay?