If you want to train a supervised learning model, you need labelled data - examples where a human has attached the correct answer. Data annotation is the process of creating those labels. It might mean drawing boxes around objects in photographs, tagging the sentiment of customer reviews, transcribing audio recordings, or marking up entities in text. The scale can be enormous: state-of-the-art image recognition models are trained on millions of labelled images, and each one required a human decision. Annotation is often outsourced to specialist companies or crowdsourcing platforms, where large teams of annotators work through tasks for modest pay. This has raised legitimate ethical concerns about working conditions and wages, particularly when annotation work is sent to lower-income countries. The quality of annotation directly determines the quality of your model. Ambiguous instructions, inconsistent labelling, and annotator fatigue all introduce errors that the model faithfully learns. Good annotation requires clear guidelines, well-designed interfaces, and ongoing calibration to ensure different annotators are making consistent decisions. It is often the most expensive and time-consuming part of building an AI system, and cutting corners here is one of the most reliable ways to end up with a model that doesn't work properly.