As we know in call centers there are certain SOP(standard operating procedure), for example few are below,

  1. Agent greeted customer
  2. Agent verified the validity of customer before providing sensitive information
  3. Agent is able to hold on conversation in decent way
  4. Agent closed conversation with nice ending notes
  5. Customer got all information what he wanted

We are planning to transcribe the call records and then from the transcribed text, we want to verify above things(Out of many) from transcribed text using natural language processing.

With the current state of AI technology is possible? If yes can someone give some guidance to approach this type of problem?


Yes it is possible to automate Callcenter conversation with Artificial Intelligence. The most famous example is IBM Watson which is an agent with interactive intelligence. That means a chatbot provides in a dialogue situation information to the customer. If this is combined with a recruiting agent to identify external experts it is even possible to automate more complex tasks which need human intervention. The problem is, that in real callcenters such a software makes no sense, because at the end the customer will speak with a computer system which has a lower service quality then a human operator.

The main problem with helpdesk agents is, that the task is way more complex then only following a standard procedure. In reality, each incoming call is equal to a new project, which is completely different and can only be solved with intensive communication between the agent and the customer, and between the agent and the back office. It is the same situation like in the secretary of a company or in Air traffic control. But suppose the idea is to automate this kind of tasks, the normal way is to use a dialogue corpus with existing recordings and build from this corpus a dialogue agent which can interact with the customer. The large dialogue corpus is used to train a model, for example an SVM model or a deeplearning network. And then, it is possible to feed a real request to this model and the neural network will output the correct reaction. Such systems works similar to the opening book in computerchess. That means if the corpus is large enough, it will provide useful answers.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.