Connect to IntercomSentiment Score
Sentiment Score is a quality metric that uses natural language processing (NLP) or AI analysis to classify the emotional tone of support conversations as positive, negative, or neutral. Unlike CSAT (which requires a customer to respond to a survey), sentiment analysis can be applied to 100% of conversations automatically, providing a far broader quality signal — including conversations that never receive a survey response. It is most powerful as a trend and outlier detection tool rather than an absolute quality measure.
Positive Sentiment Rate = (Conversations classified as positive sentiment ÷ Total conversations analyzed) × 100
Implementations vary: some output categorical labels (positive/negative/neutral), others a continuous score (0–1 or −1 to +1). Normalize to percentage of positive conversations for consistent tracking. Sentiment scores are directionally useful but not precise — current models are 70–85% accurate on support conversations.
B2B SaaS, AI-classified, all channels
Calculate positive sentiment rate
- 1Treating AI sentiment scores as ground truth — current models are 70–85% accurate on support conversations. Use as a signal for trends and outliers, not individual verdicts.
- 2Comparing sentiment across topic types — billing dispute sentiment is structurally lower than password reset sentiment regardless of agent quality.
- 3Not using sentiment to supplement CSAT on low-response conversations — sentiment analysis is most valuable for the 85–90% of conversations that never receive a CSAT survey.
- 4Ignoring agent sentiment analysis — measuring only customer sentiment misses agent frustration signals that predict burnout and attrition.