Towards Net-zero Emission: Harnessing Foundation Models for Enterprise Decarbonisation project recommendation

Abstract

As the urgency associated with climate change grows at an unprecedented pace, it has become imperative for enterprises and organizations to align their goals with sustainability values. One of the focal points for many organizations today is decarbonization, i.e., progressing towards carbon neutrality or a net zero carbon emission output resulting from their operations. Achieving net-zero emissions is a critical goal for organizations across all industries. The Net Zero target refers to completely negating the amount of greenhouse gases (GHGs) produced by reducing emissions and implementing methods to reduce the amount of GHG emitted directly or indirectly into the atmosphere. Modern regulations and economic pressures make it necessary for enterprises to exercise GHG emission accounting and disclosure and accelerate their progress towards Net Zero targets. In fact, over 20 percent of the world’s largest companies have set long term net-zero by the next decade. Unfortunately, most of them are struggling with identifying the right set of carbon reducing initiatives suitable for their organization given the budgets and sustainable goals. Ideating the right initiatives to accelerate progress in net zero requires domain expertise, experience, creativity, diversity of thought, and collaboration. This makes the process of innovating slow, dependent, and limited. Decarbonization of supply chain is an important area of study, which includes accounting of carbon footprints, identifying an inefficient process, understanding the factors which attributes to low performance and recommending the feasible intervenable actions for overall carbon reductions. This enables businesses to identify sustainability impacts across a range of attributes such as economic, environment, social and governance. It allows decision-makers to identify sustainability opportunities and prioritize reduction actions. In the contemporary literature, most of the decarbonization works revolves around carbon accounting and hotspot identification by using either qualitative [1] or quantitative approaches [2]. Reference [3] utilised Hot Spot Analysis (HAS) to integrate social and environmental dimensions along the entire value chain and to identify relevant aspects for a product specific sustainability management. Works like [4], [5] and [6] utilized LCA to evaluate carbon dioxide equivalent (CO2e) emissions and identify carbon hotspots in bio-diesel, maize silage and beef supply chains respectively. However, both HSA and LCA methodologies do not provide explainable insights of the identified hotspots. Foundation models represent a recent advancement in artificial intelligence. They involve large models trained on extensive datasets through self-supervised learning. After training, these models can be fine-tuned for different tasks, surpassing conventional machine learning and deep learning models in performance. The remarkable success of foundation models in numerous domains has prompted researchers to explore their applications in the field of climate and sustainability. In this paper, we proposed a novel prompt-tuning methodology for large language models for recommending decarbonization project based on context like type of industry, decarbonization targets, decarbonization projects already taken in past, investment amount, etc and demonstrated the efficacy of the LLM to recommend enterprise specific decarbonization projects.