A data science team at a large online retailer needs to automatically categorize customer support emails into one of 120 specific and nuanced categories (e.g., 'late delivery due to weather', 'damaged item from shipping', 'incorrect product variant shipped'). The team has access to a historical dataset of 2 million emails, each already correctly labeled by human agents. A junior team member suggests using a general-purpose, off-the-shelf large language model by providing it with a prompt that lists all 120 categories and asks it to choose the most appropriate one for each new email. Analyze the junior member's proposed solution. Is this the most effective and robust approach given the scale and nature of the problem? Justify your reasoning and propose a more suitable alternative if you believe one exists.