Customer Message Intent Classification Dataset (600 examples)

A labeled intent-classification dataset (intents.jsonl) of 600 short, synthetic customer support messages spanning SaaS/subscription and e-commerce scenarios, evenly split across 10 categories at 60 examples each: billing question, cancellation request, technical issue, feature request, compliment, complaint, account help, refund request, shipping question, and general inquiry. Each record is a flat object with an input message and a label. Messages vary in length, tone, and the specific product/plan details referenced, and are written to read as distinct, realistic messages rather than templated duplicates. All messages are original and synthetic — not copied or paraphrased from any real third-party dataset or real customer conversation — and contain no real names, phone numbers, addresses, payment-card numbers, or credential-shaped strings; any email addresses used are generic placeholders. Intended for training or fine-tuning an intent classifier or support-message router, benchmarking zero-shot/few-shot LLM classification prompts against a known balanced label set, or as a seed set for further data augmentation.