What is sentiment analysis?

Why it matters

Sentiment is the cheapest available proxy for churn risk inside a ticket queue, and it is genuinely useful for prioritisation. It is also frequently over-trusted, because polite frustration and blunt neutrality both score badly compared with what a human reader would conclude.

What to know

• Cultural and linguistic norms shift sentiment scoring, so the same message reads differently across markets.
• Terse messages from busy technical users often score negative when nothing is wrong.
• Sentiment trend across a conversation is more informative than any single message score.
• Use it to prioritise and flag, not to grade agents, or the metric becomes a source of unfairness.
• Pair it with account value, because an angry small account and an angry key account need different responses.

An example

Example: an enterprise customer writes a short, factual message about a broken export. Sentiment scores neutral. A different customer writes three friendly paragraphs about a minor annoyance and scores positive. The first one is the churn risk.

Common mistakes

• Using sentiment scores in agent performance reviews
• Prioritising on sentiment alone without account context
• Applying one sentiment model across very different languages
• Reading a single message score instead of the conversation trend

Questions