GenAI-aided Sustainable Digital Transformation: A Novel Framework and Early Results

Nowadays, rapid technological advancements play a central role in redefining the operational and strategic dynamics of businesses. Artificial Intelligence (AI), with particular emphasis on Generative AI, not only ensures greater efficiency in business processes but also holds the potential to completely revolutionize the way companies design, implement, and optimize their operations. These tools offer new opportunities to address one of the most pressing challenges of our time: the sustainability of processes. This work proposes a framework structured into three blocks to assess the contribution of Generative AI to sustainable process optimization: (i) automatic process generation through Large Language Models (LLMs), (ii) automated conversion into BPMN models, and (iii) quantitative sustainability analysis. Although the framework has been fully defined at a theoretical level, its implementation is still in progress. This work, in particular, focuses on the results achieved for the first block. Unlike traditional methods that require the involvement of domain experts, advanced Generative AI models are used to automate most (if not all) of the transformation. The study unfolds in two main phases: in the first phase, LLMs generate sustainable versions of processes from textual descriptions, following specific criteria such as carbon footprint reduction, material recycling, and energy efficiency. In the second phase, the results are evaluated using the G-Eval Framework, comparing model performances with expert-conducted analyses. For the validation of the approach, an analysis was conducted on real processes taken from the Camunda repository. Each process was provided as input to different LLMs, accompanied by a carefully designed prompt specifying the sustainability criteria to be applied. The results proved to be very promising: the Claude 3.5 Haiku model achieved the highest performance (77%), while GPT-4 Turbo scored the lowest (66%).