Presentation (CTI)

OBJECTIVES: Large language models (LLMs) are increasingly entering healthcare workflows, yet their resource use remains poorly characterized. Token consumption represents the fundamental economic unit underlying LLM inference, and token requirements vary widely across task types. Simple classification prompts may use fewer than 200 tokens, while clinical summarization or document extraction tasks can exceed 1,000 to 6,000 tokens. These ranges create uncertainty about the economic value of AI-enabled workflows. We propose a conceptual micro-costing framework, Token Based Return on Investment (Token ROI™), that integrates token use, human remainder labor, and workflow structure to clarify when AI improves efficiency, when human labor remains preferable, and how hybrid human-AI workflows influence total resource use.
METHODS: The framework adapts micro-costing principles to AI-enabled workflows at a conceptual level. It decomposes workflows into micro-tasks and describes resource inputs required for human labor and AI-supported labor. Token use is treated as a measurable and practical proxy for AI resource consumption, while acknowledging variability in token pricing, human task costs, automation probability, and error-related rework. To accommodate these uncertainties, the framework incorporates scenario-based ranges for supervision time, orchestration overhead, and error-related rework. The framework enables derivation of comparative cost and efficiency ratios, including a Token Cost Effectiveness Ratio, for evaluating human, augmented, and automated workflows.
RESULTS: Illustrative task categories demonstrate that token use can vary markedly across similar workflow types. When token variability is combined with human remainder time and supervision needs, the relative cost advantage of AI shifts substantially. These examples underscore the need for a structured approach to interpreting resource use in AI-enabled workflows.
CONCLUSIONS: Token ROI provides a conceptual foundation for integrating token consumption and hybrid workflow dynamics into micro-costing analyses. This framework supports responsible economic evaluation, procurement decisions, and governance of AI adoption in healthcare.