Automatizace procesu zpracování požadavku helpdeskem s využitím AI

In this diploma thesis, the author first described in the theoretical part the individual business entities that are linked to the process itself, which takes place at the internal employee help desk of the selected company. The following is a description of techniques, methods and tools in the field of business and process analysis, data analysis and machine learning, which the author directly used in the practical part or are related to other parts of the work, thus introducing the reader to the overall theoretical context. In the practical part of the work, the author first analyzed the current (AS-IS) state of the process in the company using business and process analysis techniques. During this analysis, the author used an iterative consulting approach with regular meetings with key business representatives. Based on this analysis, the author identified areas that could be optimized or even automated within the process. From the identified possibilities, the author chose a subset of problems to be solved, or to be implemented within the framework of this work, and subsequently proposed the future (TO-BE) state of the process and its potential benefits, which he expected in case of successful implementation and deployment of the solution. Furthermore, the author proposed metrics that can be used to measure the effectiveness of the process and on the basis of which it is possible to determine the expected difference compared to the original state. Due to the chosen area, the author applied data analysis methods to the next solution and, specifically, descriptive analytics, he extracted information from the data in the form of visualizations, which he further used in decision-making during the implementation of the technical solution and others that are key for the user for an overview of the performance of the department and people, as well as to support their decision-making, suggested for implementation as part of possible next steps. The background data for this chapter was real company data - help desk request data. Furthermore, the author created a so-called Proof of Concept solution from the field of predictive analytics - natural language processing and machine learning. After using various preprocessing techniques, the author tested various combinations of approaches, algorithms and their settings in order to train the most usable model with the highest business benefits. Such a model should be able to predict the category of the request based on its textual content and thus replace in this part of the process the human activity and the time spent reading the e-mail, evaluating it, creating a ticket in the system and subsequent categorization. The result should be a more efficient process. The author evaluated the most suitable combination of approaches based on the gradual registration of results and the evaluation of relevant metrics. In conclusion, the author deals with the real accuracy and applicability of the solution on a sample of data supplied by the business - e-mails that were processed by a person and that were not part of the training or test data set. He compared this real categorization with the predictions made by the tested models and evaluated the real success and usability. The author quantified the benefit for the business itself primarily in terms of saving human resources and also listed other impacts, among other things, as a follow-up to the previously conducted business analysis. In conclusion, the author proposed possible improvements, options for follow-on projects also within other identified areas for optimization or automation, as well as options for expansion and parameterization of the technical solution itself.