EdgeTier analyses every message and utterance from customers for emotion and sentiment. The emotions are tracked with natural-language models that are focussed on emotions that correlate with customer satisfaction - praise, gratitude, and frustration.

Language models are multilingual and industry-specific, trained specifically on customer-service focussed datasets to achieve the highest levels of accuracy possible.

Emotions Detected

Through our study of the sentences that affect Net Promoter Score (NPS) and customer satisfaction (CSAT), EdgeTier currently focus on the detection of three primary emotions in customer messages/utterances:

  1. Gratitude: Gratitude is the plainest emotion, and will consist primarily of “thank you” and “thanks”, “thank you very much” messages. Generally, the presence of more gratitude in customer interactions correlate with better customer satisfaction scores.
  2. Praise: Emphatic expressions of delight with the agent or product. These messages will be much more positive than “gratitude” messages and will generally correlate with high levels of customer satisfaction.
  3. Frustration: Frustration is the key emotion associated with poor NPS or poor CSAT scores. Messages tagged with frustration will have a very negative connotation, and frustration can be expressed with the product, service, or agent.
  4. 🆕 Customer Confusion: Confusion occurs when customers express uncertainty about a product, service, or the solutions provided by an agent. Unlike frustration, confusion isn’t inherently negative—but it highlights a lack of clarity. Confused messages point to miscommunication or gaps in customer knowledge, helping teams proactively refine processes or messaging to prevent misunderstandings.
  5. 🆕 Agent Empathy: This is a new model that looks at agent messages and detects whether the agent has responded empathetically to the customer’s communications.

Emotion detection is applied on a sentence by sentence basis in the EdgeTier system. This provides a high degree of granularity, and often, interactions will contain several different emotions throughout.

Any interaction containing a message within that has an emotion is marked with that emotion. For example, a chat that has a customer message saying “This is a joke!” will be marked as containing “frustration”.

Note that the emotions detected, along with other metrics are combined together to give, for chat interactions, an overall “score” called the Experience Score. You can read more about Experience Score on our Metrics Page.

image.png

Languages and Industries

The language models trained by EdgeTier are multilingual and will work well across all European languages. While still effective, there can be a drop in accuracy observed for Asian languages.

We have found that the sentences associated with emotions vary between different industries, and as such, EdgeTier has created industry-specific models for the travel, retail, and gaming industries.

Industry specific training greatly improves the accuracy of the output across all languages.

We have pushed these updates live to our systems. The state of go-live for these models are: