Case Study: Transforming Call Center Insights with Generative AI

Author: Tony Ojeda

In the ever-evolving landscape of customer service, understanding the nuances of customer interactions is paramount. Our recent project with a nationwide financial services provider exemplifies how leveraging Generative AI can revolutionize the way businesses extract and utilize insights from unstructured data. This case study explores our journey of transforming call transcripts into structured data, providing the client with unprecedented insights into their call center operations.

Understanding the Client’s Needs

The client, a prominent financial services provider, faced a significant challenge: their call center generated vast amounts of unstructured data in the form of call transcripts. While they had some metadata and self-reported tags from customer service representatives (CSRs), these were limited and often too general to provide actionable insights. The client sought to understand the content of these calls in greater detail—what topics were discussed, what questions were asked, what issues were raised, and how complex the calls were. This information was crucial for improving customer service, training, and operational efficiency.

The primary goal was to utilize Generative AI to extract insights from previously untapped and unstructured data sources, in the form of call transcripts, and create structured data assets that could be easily reviewed, aggregated, analyzed, and reported. This project was also an exploratory effort to demonstrate the potential of Generative AI in extracting valuable insights from previously untapped data sources.

Our Approach and Extracted Data

We began the project with a small sample of calls, aiming to extract a wide range of data points. Through an iterative process, we refined our approach, adjusted prompts, and analyzed the extracted data. Feedback sessions with the client’s teams were integral to this process, allowing us to focus on areas of interest and ensure the data’s relevance. As we expanded scope (ultimately to more than 10,000 calls), we began integrating metadata from the client’s internal systems. This allowed us to uncover insights about how other system resources impact call success.

Some of the key areas of focus included:

Topics Discussed on the Call: We extracted a list of topics discussed on each call, providing a detailed view at the individual level and enabling aggregation into higher-level topic categories.
Topic Categories: We created groupings of the extracted topics, combining AI processing with the client’s existing categories. This involved multiple rounds of processing, human review, and various methods of topic-to-category mapping.
Phrases Common to Certain Topic Categories: We identified phrases commonly present in calls dealing with specific topic categories in an effort to predict caller intentions and therefore augment call routing for complex topics. Additionally, this data point allows for better topic identification even with non AI capable tools.
Complex Question Extraction: We worked with the client to identify and extract the most interesting and complex questions from the calls so that they could be incorporated into the CSR’s training.
Call Complexity: We defined and refined what constituted a complex call, incorporating proprietary information from CSRs such as which topics are known to be complex. Each call could then be tagged with a complexity rating for further analysis.
Issues/Complaints Related to Customer Portals: We segmented calls dealing with the client’s portal and extracted relevant complaints and issues to inform portal improvements.
Insights Related to CSR Tenure: By integrating data on CSR tenure, we explored relationships between tenure, call complexity, and call sentiment. These insights encouraged conversation around improved call routing for complex calls.
CSR/Caller Followup: We extracted action items that both the Caller and CSR said they would perform during and after the call. This could then be used to better understand the actions required of both parties to resolve the issues discussed on the call.

The extracted data points were combined into a comprehensive data asset, viewable in Excel, which included the call transcript, metadata, and all extracted data points such as topics discussed, topic categories, FAQs, complexity ratings, issues/complaints, and actions taken.

Overcoming Challenges

The project was not without its challenges. One of the first hurdles we encountered was defining a complex call. We initially established some general rules to define complexity and processed a small batch of calls to review with the client. Through several rounds of feedback and iterating on our prompts we established a comprehensive set of rules customized to their needs including a list of product offerings and requests that they knew to be complex tasks.

Another challenge was finding the right balance in the level of detail for the extracted topics. The data needed to be succinct enough for easy review while detailed enough for accurate aggregation and reporting. Through multiple rounds of prompt refinement and feedback sessions, we achieved this balance.

Additionally, mapping topics to categories required a combination of AI and human expertise. We tested various methods, including machine learning models and ad hoc mapping, to ensure the categories were comprehensive yet manageable for reporting purposes.

Data Analysis and Outcomes

With the newly extracted structured data, we iteratively analyzed all of the call data to uncover insights previously unknown to the client. One of the key areas we explored was the ‘Topics Discussed on Calls,’ which allowed us to create a comprehensive view of Topic Overlap. By identifying groups of topics that are frequently brought up simultaneously, the client was able to enhance their Customer Service Representative (CSR) training programs. This insight also proved invaluable for organizing information within their internal systems, leading to improved efficiency and a more streamlined workflow.

Additionally, we extracted names and details of competing products and businesses mentioned by callers. This comparative analysis enabled the client to understand how their products stacked up against the competition. By segmenting calls based on the relationship of the caller to the account—whether they were product owners, producers, or brokers—we used GenAI to compare the types of questions each group asked. This segmentation provided a deeper understanding of caller needs, allowing the client to tailor their support strategies more effectively. For instance, identifying the caller type in the Interactive Voice Response (IVR) system before a CSR picks up the call could lead to more personalized and efficient service.

The client was also keen on improving their customer portal, so we meticulously extracted all comments and complaints related to the portal. This analysis highlighted several areas for improvement, including UI updates, feature additions, and overall UX enhancements. By addressing these pain points, the client could significantly reduce the number of calls to CSRs, thereby boosting operational efficiency and enhancing customer satisfaction.

Our collaboration with this financial services provider showcased the transformative power of Generative AI in creating structured data from unstructured call transcripts. By turning thousands of pages of raw text into a structured data asset, we provided the client with valuable insights that would have been nearly impossible to uncover manually.

The project not only informed other efforts within the client’s call center but also demonstrated the potential of Generative AI in extracting actionable insights from unstructured data. From identifying training opportunities to exploring call routing strategies, the insights gained from this project have the potential to significantly enhance the client’s customer service operations.Generative AI can bridge the gap between unstructured data and actionable insights, paving the way for more informed decision-making and improved operational efficiency in customer service environments. If you want to explore how this application or other Generative AI solutions can enhance your business, contact us today. For more insights into the world of Generative AI, follow our blog or sign up for our mailing list.