Improve customer care in Telco with Natural Language Processing

Rationalization of big data is handled by machine learning. This specific component in the field of Artificial Intelligence is very functional when it is applied in semantic analysis or, also, in voice analysis. 

The management of support tickets is an ideal testing ground for machine learning and, when the opportunity for custom development for a client arose, Lutech gladly accepted the challenge. In a context where customer relationships are extremely frequent and use different touchpoints, the primary need is often to reduce the number of tickets opened for reasons not consistent with the type of ticket itself, through new data governance.

In markets such as Telco, where ticket management is an extremely costly activity, the use of Artificial Intelligence engines helps optimize resources.

In this case, the client is one of the biggest telephone companies in the world. With a dense network of sales outlets and telecommunications infrastructure in Italy, Europe and many non-European countries, the carrier has always paid particular attention to the implementation of new technologies and to all support aspects, for end customers, partners and internal resources.

A consequence of this strong propensity to contact with various different parties is the management of tickets opened by third parties that are not compliant with their content. In other words, the person opening the ticket may, for various reasons, not be aligned with the classification logic used for them within the client company and, thus, contribute to increasing the level of complexity.

Changing the responsibility for managing a ticket can involve significant costs, especially when changes are identified after some time and, above all, there are too many tickets to be monitored manually.

The primary objective is, therefore, to analyze the content of the ticket, assess its compliance with the predefined categorization and reassign management responsibility to the team that made the mistake, limiting the loss of time and, above all, human intervention.

Aware of the client’s needs, Lutech developed a solution based on Natural Language Processing techniques in order to identify similarities between a ticket and its grouping according to the characteristics of the ticket itself. 

It was immediately evident that, as often happens in the use of Artificial Intelligence algorithms, it was necessary to turn to a development platform and a set of tools for the specific requirement to be used as a starting point. Thus, Google Cloud was identified as the environment on which to build and customize the application solution around the client’s needs. 

Lutech’s project was developed as a native cloud application, implemented on Google Cloud Platform. It consists of two main phases of analysis: identification of text properties and grouping of tickets with similar characteristics. 

The same features were used for the creation of groups of tickets whose number has been determined automatically, again based on text properties. 

Subsequently, the customer’s business team provided labels to define clusters, allowing the definition of a real classifier, whose accuracy increases as users provide feedback on the proposed classification (machine learning). 

In particular, the application developed by Lutech provided a much faster compliance assessment, on the order of a tenth of a second; the possibility of continuous learning; and the ability to configure custom notifications based on the type of inconsistency found.

To date, the application system implemented by Lutech has allowed the time elapsing between ticket opening and its compliance analysis to be virtually eliminated, bringing it down to around a tenth of a second, while in the past this could take as long as a few days and depended heavily on the number of requests and the management of the resources in charge. 

Today, the system immediately identifies the tickets with an invalid format content and suggests to the operator which ones are suspect, speeding up the whole analysis and processing process, with a significant improvement in customer satisfaction.