What is sentiment analysis?
Sentiment analysis is a form of natural language understanding that looks at how something was said rather than what it was about. It reads a customer’s words and estimates their emotional tone. The simplest version labels a conversation, sentence or email as positive, neutral or negative. More detailed versions look for specific emotions such as frustration, confusion or relief, or score how strongly the feeling is expressed.
In voice channels, sentiment is usually worked out from the transcript produced by automatic speech recognition. Some systems also use acoustic cues like pace, volume and pitch.
How contact centres use sentiment analysis
In real time
- Escalation. If a customer talking to an AI voice assistant or chatbot becomes frustrated, a sentiment signal can trigger an AI-to-human handover before the conversation gets worse.
- Prioritisation. Angry or distressed emails can be moved up the queue.
- Agent support. A live indicator can alert a team leader that a call is going badly so they can step in.
After the interaction
- Trend tracking. Sentiment across all contacts, broken down by reason for contact, shows which issues upset customers most.
- Root cause. A sudden drop in sentiment on one topic can reveal a new billing problem, process failure or service change that isn’t landing well.
- Quality and coaching. Within quality assurance, sentiment at the start and end of calls shows whether agents are turning conversations around.
This is where sentiment analysis is most often used, as part of speech analytics.
Sentiment vs satisfaction
Sentiment analysis and customer satisfaction scores measure related but different things. CSAT asks customers directly, but only some answer, and they tend to be the very happy or very unhappy. Sentiment analysis covers every analysed conversation but infers feeling from language. A customer can sound negative because of the situation (their power is out) while being entirely happy with how the call was handled. Good reporting keeps those apart, for example by comparing sentiment at the start of a call with sentiment at the end.
Limits to keep in mind
- Sarcasm and understatement. “Oh great, another outage” reads as positive to a naive model.
- Cultural and personal differences. People express frustration very differently. Some are polite even when very unhappy.
- Topic vs tone. Calls about bereavement, debt or outages will often score negative regardless of how well they go.
- Over-reliance. Sentiment should prompt action by a person, not make decisions about a customer on its own.
Used with those limits in mind, sentiment analysis is one of the most practical ways to hear the mood of your customers across far more conversations than any team could listen to.