Automating climate commitment analysis

30% of top global companies lack structured climate data, and for most, the costs of manual data structuring remain prohibitively high. The market needs innovation, especially as climate reporting becomes mandatory. BlackRock's latest research paper introduces an AI system for extracting climate commitments from corporate documents, achieving 95-100% accuracy. The technical approach is elegant in its simplicity, breaking documents into 80-word chunks, using RoBERTa to identify relevant sections, and leveraging LLMs for precise data extraction. An important addition is their multi-stage validation system that includes hallucination checks and semantic deduplication - essential features for production-grade AI systems. While initially tested with Google's PaLM2, the researchers demonstrate comparable performance with GPT-3.5 and GPT-4, suggesting a robust, platform-agnostic solution. However, some limitations: the validation dataset, comprising only 46 companies, is relatively small, and the system's performance on non-standard reporting formats remains untested. As reporting requirements evolve globally, maintaining the system's accuracy will require ongoing refinement. With over 30% of the top 1,000 firms lacking structured climate data, automating climate commitment analysis should improve market transparency while reducing costs, and reducing the barrier to entry. Excellent (and practical) work by Aditya Dave, Mengchen Z., Dapeng Hu, and Sachin Tiwari from BlackRock. PS: HAPPY NEW YEAR! ❤️

🆕 NEW PAPER PUBLISHED: Using AI to assess corporate climate transition disclosures 💚 I'm happy to announce that our paper was published in #Environmental #Research #Communications. The paper presents a blueprint for employing #Large #Language #Models (#LLMs) to analyze corporate climate disclosures in detail and with experts in the loop. We proceed in three steps: 1️⃣ #Framework for #Transition #Plan #Assessments: We build the common ground of 28 transition plan frameworks to create 64 key indicators for evaluating companies' efforts to transition towards net zero. 2️⃣ #Expert #Validation of the corresponding #LLM #Tool: We build an LLM-based tool that can automatically and efficiently analyze disclosures (such as sustainability reports). We validate and improve the tool with domain experts from 26 different institutions, including financial regulators, investors, and NGOs. 3️⃣ #Analysis of the #Most #Carbon #Intensive #Companies: Analysing the highest-emitting companies in the world, we find a gap between "talk" (target setting) and "walk" (strategy implementation). Companies with more disclosures tend to have lower emissions. Similar to commitments towards science-based targets, larger emitters remain more intransparent. 🔗 The paper is here: https://lnkd.in/dJkCjFHV 🌟 The best thing? It is all #open-#source and publicly available. You can use the tool yourself, work with it, and improve it. I even wrote #tutorials for newcomers to the field. So really nothing is stopping you from trying it out. 🔗 Check it out here: https://lnkd.in/dgFPqydB This paper is only possible through a great interdisciplinary collaboration with Chiara Colesanti Senni, Julia Bingler, Jingwei Ni, and Markus Leippold! Feel free to use and provide feedback! University of Zurich | ETH Zürich | University of Oxford #climate #change #NLP

Krista Tukiainen is building an AI engine to scale adaptation finance. (and major asset managers are taking notice) Krista started her career in sustainable finance. But after nearly a decade working with green bond standards, she saw a problem: Everyone was making climate commitments. But no one could agree on what counted — or how to track it. So in 2022, she co-founded ClimateAligned to fix that. Her team's mission: → Automate the messy, manual work of parsing disclosures → Make sustainable finance auditable, scalable, and transparent The tech? A self-serve AI platform that can run bespoke assessments on all kinds of public filings, surfacing decision-useful data for investors, corporates, consultants -- and journalists! Krista's team calls it brief.green I'm using the tool to assess how aligned the world's top 50 banks are with adaptation indicators. One asset manager may use ClimateAligned to create a whole new adaptation-focused fixed income fund. 🎧 Learn more by listening to the full conversation on the latest installment of the Climate Proof podcast: https://lnkd.in/e4veDVde 📩 Subscribe for more deep dives on adaptation finance, tech, and policy: https://lnkd.in/eFDX8kWn