GENERATIVE MODELS FOR CREATING RESONANT COMMUNICATION SCENARIOS IN CRM SYSTEMS

Abstract

The article presents a modular CRM architecture that combines emotional-cognitive client analysis with dynamic generation of personalized communication scenarios using large language models. A key element is the structured prompt mechanism, which includes role instructions, dialogue history, a description of the client’s state, and business rules. The aim of the study is to develop and experimentally evaluate an approach to scenario generation that simultaneously accounts for the user’s emotional and cognitive parameters to enhance the resonance of responses and the quality of interaction.

The client’s state is represented as two normalized vectors (emotional and cognitive) which allow for a compact description of current reactions and intentions. Based on these vectors, a dynamic structured prompt is formed and used to generate relevant alternative response scenarios. Each scenario is evaluated through a resonance function, and the optimal option is selected by a strategy selector that considers resonance and prior effectiveness. The balance between emotional and cognitive components is dynamically adjusted by an RL agent.

The work formalizes a two-component (emotional and cognitive) representation of the client’s state for generative personalization, introduces the resonance function as a metric of scenario relevance, and for the first time combines dynamic prompt formation with emotional-cognitive client analysis to build high-resonance communication scenarios.

System effectiveness was evaluated in a simulated environment based on 500 dialogue sessions with variable emotional-cognitive client profiles. Comparison with a baseline model demonstrated significant improvements in key metrics: the average resonance score increased from 0.62 to 0.71, the share of high-resonance scenarios rose from 28% to 43%, and the number of low-resonance scenarios was halved – from 22% to 11%. The results confirm that dynamically generated scenarios provide higher personalization, behavioral stability, and potential conversion growth.

The proposed model demonstrates the effectiveness of generative algorithms in CRM systems capable of adapting the style and content of communication to the client’s current emotional-cognitive state in real time. The mechanism of dynamic prompting and resonance-based scenario selection can be integrated into chatbots, contact centers, and service platforms, improving the quality of automated responses and reducing the share of irrelevant messages without changes to business logic.