the cognistx blog

Call Summarization: an AI Solution for Customer Service Providers

August 4, 2020
By
Jagriti Pandey

For any Call Center, the performance metric is more focused on speed and subjective elements than the depth of problem-solving. But, according to statistics from Microsoft’s State of Global Service:

  • According to 36% of survey respondents, the most frustrating aspect of a poor customer service experience is when an agent lacks the knowledge or ability to solve the customer’s issue. Whereas, 31% say it is having to repeat or provide their data multiple times which incorporates expressions like:
  • Transferring Ineffectively - “Ahh! No, Not Again!”
  • Forcing Repeat Information - “Blah! So irritating. Kill Me!”
  • Menu Mazes - “What the heck! Where is the option I am looking for!”

There has been an unprecedented rise in omnichannel services. As mentioned in biz30, businesses rely heavily on customer surveys (77%) and call monitoring (64%) to record customer feedback. However, there has been an unprecedented rise in social media listening (45%) over the past few years.

A phone call may be a traditional and cozy way to reach out for not so tech-savvy people or even otherwise, but the other channels such as social media, live chat, and email will continue to see the surge in their usage.





Now imagine how much information a call center or a customer satisfaction monitoring application has! Making sense of this unstructured information and utilizing it 100% to optimize your call center can be a far-fetched dream without a streamlined process.

If the evaluations have to be done by human reviewers, it will first take long working hours to conclude how to utilize it to optimize the customer service. Be it ineffective transfer of calls or forcing the customer to repeat their problem - summarization can help to optimize the call rerouting and save the customer from repeating oneself by providing a prior summary to each new representative of the call center while the call reroutes to a new representative.

The value of the data by converting from speech to text and then summarizing it varies concerning the purpose of the summary. One of the use cases can be to create a knowledge-work environment like a T-Mobile contact center, as mentioned in HBR. Summarization from various omnichannel sources and the calls can provide crisp information about ongoing issues via knowledge graphs or other tools. This ensures the service providers are in touch with current customer pain points and changing business needs and trends. Once you have this information in knowledge graphs or similar data structures, you can go ahead and create a small group of Subject Matter Experts (SMEs) in this field to solve customer issues much more swiftly than it was being taken care of, or one can get to the root cause to eliminate certain types of customer complaints or product question calls.

For one of our prototypes, Cognistx focused on address formatting so that after the user has provided their address via speech, this system formats the address to make it precise. This application helped to curtail long calls to find out that the address is not serviceable.


Fig 1. The Address Voice Analysis showing the formatted address

Meanwhile, another use case where Cognistx provided Agent Performance Analysis, Call Transcript Analysis, and Call Quality Management/Analysis via call summarization. The Call Summarization helped the supervisors from manually rating and ranking all the calls. Since it was humanly not possible to rate and rank each call, therefore, the calls were monitored and rated at random. The long and arduous feedback loop was curtailed by having a summarized report for calls.


Fig2. Customer Transcript Analysis

In contrast, if the focus is towards the call segregation and categorization expertise, or to understand the overall call sentiments, then the call summarization would include other aspects of analysis. Such analysis can consist of insights about the bucket a call from the customer can possibly fall under via topic modeling, emotion analysis of the call via sentimental analysis, or feedback summarization via call summarization. So, the value a call summarization provides us depends on the requirement for a particular use case.

There are various factors - from the conversation time duration to emotion recognition - the basic ones: anger, frustration, empathy, and satisfaction, that influence the evaluation of the call summarization.

There is no straightforward evaluation when it comes to call summarizations. We believe that call summarization should be in line with the objective that enables the organization to utilize AI to its fullest potential.

Recent Posts