A customer service company uses a large language model to analyze support chat transcripts. The model was originally trained on text segments with a maximum length of 4,096 tokens and performs well on short conversations. However, its performance degrades significantly when analyzing complex, multi-turn support cases that often exceed 8,000 tokens. You are tasked with improving the model's performance on these longer transcripts. Describe the most critical characteristic of the data you would select for a targeted fine-tuning process to address this specific problem.