Success Story - en
NLP Contact Centers Engineering Division

Tagging dialogues of the Contact Center

Description:  automation of checking the effectiveness of advertising campaigns and evaluating the activities of advertising agencies and evaluating the conversion rate in the customer acquisition funnel at the stage of processing a phone call from a client to a contact centre operator requires an assessment of the probability of one of the specified events in the conversation between the contact centre operator and the client, based on a recording of a phone conversation presented as text. These events are the target customer actions and intents of the call.

Context:  the company launches an advertising campaign that aims to create a flow of requests to the contact centre on a specific topic (for example, to sign up for a test drive of a car). The effectiveness of such campaigns is difficult to measure since you need to understand the content of each dialogue. Manual dialogue markup was an inefficient process.

Decision:  it was suggested to build a dialogue tagging model based on the created annotation.

Results:
The solution is built and adapted for:
  • Car dealer companies;
  • Real estate sales companies;
  • Companies-medical centres.

Integration into the BI cloud platform automation of CRM management for the customer's clients:
  • CRM completion is 40% more and 20% more accurate;
  • Automation of measuring the level of conversion in the channels to attract;
  • Reduce labour costs by 60%;

To build the model, we used:
  • Dialogues between contact centre operators and the client divided by replicas;
  • Information about the presence of events in the dialogues;
  • Expert knowledge of the essence of events.

Simulation result:
  • Model for predicting the probability of an event;
  • Analytics of the dialogue data set;
  • Ad campaign conversion Analytics.

Customer: Contact Center, Telecom

Technology stack: text-preprocessing models, classification models, language representation models, Flask, Python, PyTorch.