<img src="//bat.bing.com/action/0?ti=5739614&amp;Ver=2" height="0" width="0" style="display:none; visibility: hidden;">

Artificial intelligence is only as intelligent as the architects. In fact, AI can learn a lot from us. Especially when it comes to customer service.

woman-with-headset-in-call-centerCornell University recently published a study by our VP of Data Science and Innovation Sony SungChu and Research Scientist Yanan Jia discussing the importance of using sentiment analysis when programming AI to help it replicate and understand human language. By “studying” the formation and responses of well-trained customer service reps, AI can be developed in a more empathetic way and, in turn, help make the interaction between humans and AI more fluid and more impactful.

I sat down with Sony to get a better idea of what this means for the benefits industry.

How does sentiment analysis help improve AI capabilities?

Today, AI alone is generally not enough to account for the entire set of possibilities that can come up in a customer service interaction.

Sentiment analysis can help identify delicate interactions where a human agent might be needed. It can also help de-escalate tense situations by helping automate notifications to call center supervisors so that they can intervene early in a call, when necessary. We also see application of sentiment analysis in identifying when a virtual assistant requires calibration.

For example, we use sentiment analysis to identify when Sofia or a live agent has had a sub-par interaction, this allows us to identify new training opportunities to handle more complex interactions or refine existing ones regardless of whether it is a live agent or virtual assistant taking the call.

How did you get started on this topic and why is it important?

While conducting our literature review of existing publications and research, we found that most of the studies that have been done on sentiment in conversations were done in controlled environments without significant use of real-life, multi-party calls.

Our study is done entirely on data we’ve collected over the years through Businessolver’s benefits call centers. This is advantageous to a study because we are able to analyze and overcome a wide-range of challenges including picking out the right signals from noisy inputs. For example, environmental noises, variations in word pronunciations and intonations, overlapping dialogue, etc. In overcoming some of these challenges, we’ve built a method that is capable of analyzing multi-party sentiments in real world calls.

In addition to representing real world scenarios, our method presents a two-pronged approach where we not only study the textual information offered in a conversation, but we also factor in the physical characteristics, such as the fluctuation of energetic levels across frequency ranges, of the actual audio recording.

How can sentiment analysis make AI more empathetic when in a customer service role?

There are many factors that can contribute to the sentiment of a person in a call. If the sentiment is negative, it requires AI to take additional steps to diffuse the situation and provide options that lead to a more positive experience. Given that interactions are not always linear, meaning they can flow from positive to neutral to negative in any order, the first step towards an empathetic interaction is to correctly identify the sentiment and offer responses that befit the caller’s sentiment.

The next step is to make sure your AI has enough information and capabilities built in to either help the caller or find a human to enhance the call experience. The strength of AI is that the more interactions it is exposed to, the more it can learn to be empathetic.

How are we using sentiment analysis to improve our own AI personal benefits assistant?

We use sentiment analysis to identify opportunities where Sofia can improve. This can be a complex process since there are many factors that can contribute to a person’s sentiment. For example, a caller may ask why a dependent was not verified. Even though Sofia gives the correct reason, they may not be satisfied, or they may disagree with the reason entirely.

Depending on the situation, Sofia could be seen as working correctly, since she pulled the right information, however, without sentiment analysis in place, we miss the opportunity to identify and improve the service based on the immediate feedback that is being provided by the caller. We also use sentiment analysis to identify where Businessolver as a whole can make changes based on the reaction of callers to our products and services. This can range from measuring the overall satisfaction with the support teams to identifying confusion over a feature or program that gets rolled out.

Want to learn more about how artificial intelligence can go hand in hand with empathy? Check out our e-book below.

AI and empathy

View all Posts by Tom Staroszczyk