There are plenty of pre-trained models available, that could cover most or your needs. However, there might be some situations, when base model need to be further calibrated in order to serve well, some examples:

• Adapting a model to a specific task. Sometimes, the performance of general-purpose models is not enough for you particular problem.
• Fight with overfitting. When a model is trained on a limited dataset, it might perform poorly on unfamiliar data. Further fine-tuning can help overcome this problem.
• Knowledge transfer. - You can use another model, which is already trained to perform well on a specific task, to transfer its knowledge to your model.