Tethr and Awaken Intelligence join forces as Creovai
Tethr and Awaken Intelligence are becoming Creovai, bringing together best-in-class conversation analytics and real-time agent assistance.
Robert Beasley
June 3, 2024
Tom Shepherd
May 22, 2024
Imagine saying this to a customer service agent in your contact center: “Our sentiment analysis reporting showed you had a lot of calls with negative customer sentiment this week. Can you work on making the customers you talk to feel better about their interactions with us?”
The customer service agent couldn’t act on that request. They don’t have enough information about what they’re doing that could be negatively impacting their customers’ attitudes or what they could do to improve their customers’ experience.
In that same vein, customer experience leaders can’t make meaningful decisions based only on the information that customers expressed positive, negative, or neutral sentiments about their interactions with the business. Sentiment analysis must go beyond tracking what customers feel and drill into the “why.” When CX leaders can answer those “why” questions, they can identify the right initiatives to increase customer loyalty, reduce churn risk, and grow customer lifetime value.
However, most sentiment analysis tools for contact centers don’t deliver this level of nuance. They analyze the words customers use in conversations to simply tell you whether the customer’s sentiment is positive, negative, or neutral. There is value to segmenting customer interactions this way–it’s a useful starting place that allows you to look at big-picture trends in sentiment over time. However, for sentiment analysis to be actionable, business leaders need to understand the emotions their customers–and agents–express, as well as the factors impacting those emotional states.
With this in mind, Tethr developed a sentiment analysis model that provides overall sentiment scores for customers and agents and identifies 28 distinct emotions based on the words and phrases that appear in interactions. This gives business leaders a deeper understanding of how their agents and customers feel, allowing them to make targeted improvements to the agent and customer experience.
Tethr’s sentiment analysis uses machine learning, a type of AI, to analyze words in conversation transcripts and identify the emotions associated with them. Machine learning models must be trained on large data sets to identify mathematical relationships between words. In this case, we trained our sentiment analysis model on a large set of example phrases labeled with associated emotions. For example, “I’m really mad,” “I’m ticked off,” and “this is infuriating” are a (small) sample of phrases that would be associated with anger.
We trained our sentiment analysis model to correctly identify all the emotions associated with the training phrases. We then introduced a new set of test phrases and tuned the model to accurately identify the emotions expressed in the new phrases based on probability. For instance, the phrase “I’m very angry” would be scored with a high probability of the emotion anger.
Note: It’s important to keep training data and test data separate when building machine learning models. We use our test data to evaluate the trained model–we never introduce the test data during the training process.
Using this approach, we developed a model that provides overall sentiment scores (positive, negative, or neutral) for customers and agents and tags 28 distinct emotions where they appear in transcripts. We presented the model to a group of opt-in, early-adopter customers for feedback. We used their input to retune the model, further improving its accuracy. Going forward, all customers will be able to share feedback in the Tethr platform about whether they think sentiment scores and identified emotions are accurate or inaccurate, allowing the model to continue learning and further increasing its accuracy.
Traditional sentiment analysis models detect three sentiment states: positive, negative, and neutral. (The scores are sometimes simplified to just a smiley face or frowny face.) So why did we develop a model that detects 28 distinct emotions for both customers and agents?
Emotions are complex, and just knowing that a conversation was positive, negative, or neutral doesn’t give you the full picture. For example, a customer expressing disappointment about a late shipment differs from a customer expressing anger about a late shipment. An agent would likely handle each situation differently, and your business would likely make different decisions based on insights into different emotions.
Knowing the distinct emotions that occurred in a conversation–and seeing where they appear in the transcript–helps you understand not just how your customers felt, but why they felt that way, and what your business can do to improve their experience. You can even track changing emotional states across conversations to see what your company is doing well or could improve. For example, you could look at conversations in which a customer started out frustrated but ended up expressing gratitude. This would help you see what your agents are doing well or what offers resonate with your customers.
Sentiment analysis often focuses on the customer, but there are two parties in every contact center interaction: the customer and the agent. The emotions the customer expresses are likely to be different–or differently motivated–than emotions expressed by the agent. It’s important to break out separate sentiment scores for the customer and agent to make this distinction.
Tracking separate sentiment scores and sets of emotions for the agent and customer can help you understand how the two parties are reacting to each other. Is the agent becoming angry when the customer becomes angry, or remaining calm and confident? Is the agent expressing nervousness when the customer expresses disappointment? Understanding the agent’s emotions helps you track how the agent impacts the customer experience. You can pinpoint opportunities for improvement and coach your agents accordingly.
Uncovering insights into customer and agent emotions can help your business make informed decisions across multiple functional areas. This information helps you improve agent performance, product quality, contact center operations, marketing and sales messaging, and more. All of this is in service of the customer experience–and improving your CX has a tangible, positive impact on business growth. When customers have good experiences with your business, they are more likely to recommend your business, pay premium prices, and keep buying from you.
Tethr’s sentiment analysis model is built to help businesses get to the heart of their customer experience so they can meet–and exceed–their customers’ needs. When your business makes data-backed decisions to drive positive customer experiences, it’s a win for your customers, agents, and the business as a whole.