Sentiment Analysis
Automated classification of the emotional tone of a customer message, usually positive, neutral, or negative.
Sentiment analysis reads the language of a conversation and labels its emotional charge, at the message or ticket level. Teams use it to triage angry customers faster, flag at-risk conversations, and get a coverage signal on interactions that never got surveyed.
Output is a label or a score; quality is judged against human agreement on a sample.
What it hides: sentiment models stumble on sarcasm, terse professionalism, and domain-specific phrasing, and a polite message can mask real anger (or vice versa). Treated as ground truth it misleads; treated as a prioritisation hint that a human confirms, it earns its keep. It is a proxy for feeling, not a measurement of it.