Využití AI agentů v bankovnictví

This thesis investigates the potential for the deployment of AI agents in mobile and online banking. The theoretical part defines the concepts of artificial intelligence, large language models, and AI agents, outlines the current state of AI adoption in financial services, and analyses the relevant regulatory framework. On this basis, the concept of the agent is established, and two research hypotheses are formulated. The practical part presents a functional prototype of an agent developed in a simulated environment using a synthetic dataset and supplemented by insights obtained from a semi-structured interview with an expert from Czech banking practice. The evaluation of the prototype using the metrics of clicks-to-action, time-to-action, and intent recognition accuracy shows that, for more complex operations, the agent reduces the average number of steps by 68 % and the average time required for completion by 28 %, whereas it does not provide added value for simple operations. The agent achieved 100 % intent recognition accuracy within a controlled test set. The results indicate that such AI agents do not constitute a replacement for the traditional graphical user interface; rather, they are best used as a complementary solution, particularly suitable for operations of medium and higher complexity and for facilitating access to functions that are otherwise difficult to locate.