Conversational labeler

Improving customer service in call centers by structuring recordings from past calls with NLP and NLU


Call centers can provide a wealth of information about customers just through daily calls and trouble-shooting. But accessing and making use of this information is challenging. Though most call center calls are recorded, making sense of the information is difficult because it requires someone to listen to the recordings, which takes enormous time and resources.

One client wanted a system to glean significant information from these calls and use it to improve customer service and inform new product and service offerings.


Cognition’s conversational labeler module transcribes conversations, extracts information and stores it in a structured format so it can be analyzed and exploited. It uses natural language processing (NLP) and natural language understanding (NLU) technologies to extract structured information out of the voice calls. This information is stored in a database that the client can later use to provide better service when customers call, estimate the propensity of customers to buy a new product or service, and to design personalized campaigns and offers.

Cognition’s team analyzed a sample of the calls and discussed with the client which categories, topics, entities, keywords, expressions and collocations to extract and analyze, and coordinated on database structure. This information was used to adapt the conversational labeler to the specific call center domain.


The client now has a customer database that allows them to launch marketing campaigns customized to target specific groups of customers — previously they could only run campaigns that targeted broad customer groups. The client is also now able to provide better customer service through human agents who now have access to information shared by customers in previous calls.