Evaluating Multi-Step LLM-Generated Content: Why Customer Journeys Require Structural Metrics

Evaluating multi-step customer journeys generated by large language models is challenging with current methods that focus on style over structure. This article introduces a new framework with three metrics: Continuity, Deepening, and Progression (CDP). The approach involves first building a hierarchical content taxonomy from semantic embeddings and then mapping the customer journey onto this structure. Continuity measures thematic coherence between steps, Deepening tracks the move from general to specific content, and Progression evaluates movement through predefined business stages.