Case study: Natural language processing for call centres

AI-driven NLP models implemented in some areas of an organisation's workflow can significantly improve the quality of time spent writing and reading e-mails and reduce response time. That leads to higher productivity, advancing customer service, and saving the company's funds.

Business case:  

Call centres have to deal with plenty of text and audio information, and they spend too much time processing it. By integrating machine learning and NLP into the procedure, you can predict customer behaviour and detect sales potential while saving time and costs.

Use case 1: E-mails classification and prioritisation

Challenge: The company receives about 30 000 e-mails monthly. Therefore, it needs 10-11 employees working full-time in the call centre to process these e-mails.

Goal: To automate the indexation process as much as possible using machine learning and natural language processing/text mining methods. This will consequently free up the capacities of the call centre operators for more value-added activities.

Approach: The solution is based on the existing platform, which contains:  

• backend for natural language processing, including lemmatisation, entity and topic detection, category classification
• web-based front-end for setting up the workflow for manual classifying of missing or incorrect arrangement and visualising

We tailored the solution specifically to the needs of the given problem and customised the interface for integration with the call centre system.

Use case 2: Understanding the content of the calls and finding sales signals

Challenge: The company aims to understand the content of the inbound and outbound calls and explain what specifically makes the clients come to a branch office and buy their product. Additionally, it is possible to enhance the call scripts using the collected information.

Goal: To identify the factors during the outbound call that affect the client's actions, make them visit or not visit the branch office, and purchase or not purchase the product.

Approach: Processing transcripts of the calls, information about visits and purchases, and the customer context, such as what products they bought or their demographics.

Stored information allowed us to build a model based on text flags such as frequency, term frequency-inverse document frequency, word2vec algorithm, n-grams. The model also uses customer data to predict the probability of purchasing a product based on a call inviting to a branch office.

Benefits: 

• Understanding the customers improves KPIs.
• Automatisation of the classification and information extraction process helps improve the call scripts.
• Decreasing the number of call centre operators needed allows the employees to perform more value-added activities.
• Identifying the factors that affect customers' decisions increases the success rate of calls inviting to the branch office.