What is Sentiment?

All conversations that are imported into Evaluagent via an integration as a full fetch ticket will be analysed for Sentiment.

Sentiment can be described as the analysis of conversations to gauge polarity and emotion in a conversation with reference to context and inter-dependency. In simple terms, this allows us to better understand the positivity or negativity of parts of a conversation or the conversation as a whole.

The call analysis will cover:

• Customer Sentiment - a score between 0% (Very Negative) to 100% (Very Positive) that aims to reflect how the customer ‘felt’ (sentiment) during an analysed conversation.
  
  
• Prolonged Sentiment - Whether or not prolonged positive or negative sentiment is identified within the analysed conversation.

The sentiment score is shown as a percentage, where 0% is very negative and 100% is very positive.

Prolonged sentiment is flagged if 3 consecutive sections of the conversation (utterances) are identified as strongly positive or negative.

What metrics are available?

The analysis of sentiment provides the following metrics:

• Overall Customer Sentiment
• Overall Agent Sentiment
• Prolonged Negative Sentiment
• Prolonged Positive Sentiment

Requirements:

Sentiment relies on being able to analyse a conversation to identify periods of positive or negative sentiment.  Sentiment analysis only applies if you use full fetch tickets with your selected integration with the Evaluagent platform. 

The quality of the voice recording is also very important to ensure the best sentiment analysis. Good quality stereo call recordings with speaker separation are perfect.

Viewing a conversation's Sentiment

Each conversation that has been analysed will be tagged with the sentiment metrics that the EvaluAgent platform has identified.

Navigate to the 'Evaluagent Insights' tab for any conversation within the imported contacts page and find the sections called 'Sentiment Insights'. Here, you can see the overall agent and customer sentiment scores and if any prolonged sentiment (positive or negative) has been identified.

The sentiment score is shown as a percentage, where 0% is very negative and 100% is very positive.

If prolonged sentiment is identified, then this will be highlighted under the heading 'Prolonged periods of positive or negative sentiment' as shown in the screenshot above.

If no prolonged sentiment is detected, then this will be flagged as per the screenshot below.

For more information about how the Sentiment score is derived, please visit the FAQs section below.

Finding individual instances of sentiment or prolonged sentiment

Clicking on the yellow highlighted Prolonged (Negative or Positive) text will show you where the instance is in the text transcript. You can also navigate between these occurrences by clicking 'Next'.

The screenshot below shows Prolonged Sentiment being highlighted in the call transcription (note in live instances of this feature, you will see the actual words of the transcript, the screenshot below is a visual depiction).

Using sentiment as filters to discover conversations

Within the Imported Contacts view, you can access all of the available sentiment metrics as filters. This enables you to use the sentiment metric filters to discover specific types of conversations that you might wish to explore further or carry out detailed evaluations on. For example:

• identifying all conversations with high or low customer sentiment scores.
• identifying all conversations that have prolonged periods of negative sentiment.

To find these filters:

1. Navigate to the Imported Contacts page,
2. Select 'Create new filter',
3. Choose Sentiment Insight Fields,
4. Select the sentiment metric you wish to use in your filter,
5. Enter the settings you need for the specific sentiment metric filter you have selected.

This filter will then find any conversations that match the criteria you have specified.

NOTE: The Agent Sentiment score, although captured as a metric within the Evaluagent platform, is not yet available as a filter. This will be coming soon.

Reporting on Sentiment

Both Customer Sentiment Score and Prolonged Sentiment (Positive or Negative) metrics can be used with your SmartView Dashboard to create sentiment focused dashboard widgets to help you track these critical performance indicators. 

Frequently Asked Questions

How does Sentiment Analysis work "under the hood" ?

We generate our sentiment results using a large language model, asking the generativeAI to steps into the shoes of either the customers or agent, and based on that perspective, form its own opinion and then provide a sentiment score based on the transcription that we send it.

This approach uses the same underlying technology and approach as the xNPS feature.

The sentiment score ranges from 0% - 100% where 0% indicates an extremely negative sentiment, and 100% represents an extremely positive sentiment.

Unlike historic approaches to Sentiment Analysis, using Natural Language Processing, which looked at individual snippets of the conversation (utterances), and then calculated a combined score for the entire conversation, which led to lots of problems due to a lack of context, our sentiment analysis technology attempts to understand the whole context of the conversation before making its judgement and giving an overall view - which is expressed as a score.

Managing Disagreements over Sentiment Scores

Sentiment analysis has been a topic of debate in the contact center industry for over a decade. While the technology can offer value in prioritizing interactions for deeper evaluation, there can be disagreements with the assigned sentiment scores.

The reason for this is that sentiment analysis, by its very nature, aims to quantify an inherently subjective emotion. People often have their own biases and expectations about what "positive" sentiment looks like, based on their personal experiences and how they use language. Something that one person considers positive may be interpreted as negative by someone else. Previously this has been exaggerated by the limitations of Natural Language Processing (NLP) also. 

The use of advanced language models has improved sentiment analysis, but it's important to recognize that it is just one tool in the arsenal for identifying high-risk/high-value conversations and understanding emotions at scale. Sentiment scores should be used as a starting point for further investigation, not taken as absolute truth.

In our experience, it's common for stakeholders to disagree with individual results from time to time. When they do so, we've found a conversation is the best approach to building trust. These conversations can focus on: 

• Demonstrating how the technology is being used as a guiding indicator to find High Risk / High-Value conversations that are then pushed through a more detailed Quality Assurance process.
• Recognising and reaffirming that sentiment analysis is an imperfect science and that there will always be some degree of subjectivity and disagreement.
• Reminding stakeholders that sentiment analysis is just one data point among many that should be used to inform decision-making. The scores should be considered alongside other metrics and business objectives.