Better customer experience with sentiment analysis

Automated customer service phone calls and chatbots are becoming increasingly easy to interact with. However, the frustration associated with bad experiences can have a significant impact on customer retention. Deep learning models such as Natural Language Processing (NLP) are ideal for gaining insight into the user experience in customer service interactions.

Ever felt customer service could be more smooth?

Want to replicate this use case in your business? Get in touch

02/ What can be achieved?

Deep learning can predict the sentiment of the customer during customer service interactions which is crucial for improving the customer experience. An example of what can be done with this information is to automatically detect when to switch a customer from an automated customer service system to a human operator.

03/ The problem

Automated customer service systems are great for reducing phone queues efficiently solving simple problems but when the customer is stressed with a large problem, most people want to speak to a human operator. The traditional solution to this would be to invest in a more sophisticated automated service or invest in more human operators. With deep learning, these costs can be avoided by effectively allocating the human operators’ time to problems where human interaction is necessary.

04/ Opportunity for deep learning

NLP models are able to use the text a customer writes in a chatbot to pick up on specific things the customer mentions as well as detect sentiments like frustration. Customer service phone calls can also be analysed using speech-to-text technology to analyze what the customer is saying, in combination with analyzing the audio directly to understand how the customer is saying it.

05/ How is the AI model implemented?

The model performs automatic real-time analysis of audio and/or text data from a customer service interaction. Once a threshold level is reached, the customer will be transferred to a human operator. Alternatively, the insight gained about exactly the point in the interaction can be used to analyze what parts of the interaction are causing frustration to improve the customer service experience.

06/ Data requirements

The model needs to be trained on the text or audio from historical examples of successful and unsuccessful automated customer service interactions.

Use our BERT snippet to create a model capable of detecting sentiment from text in a matter of seconds. Do the tutorial linked below to learn about the steps in this process.

07/ Test it yourself

Use the app below to test the functionality of a model we built using the Peltarion platform to detect positivity in text data collected from Twitter. Type a sentence into the box below and click the button to see how positive the model thinks your text is from 0-1.

08/ Curious to learn more?

This use case is a text classification problem. To see how to build a model that can do this, follow the Movie review feelings tutorial.

To see how to build a model which could detect a person’s tone using only the audio, check out a similar problem in the Build your own music critic tutorial.