Why we use Machine learning
We are proud to leverage cutting-edge machine learning technology to deliver powerful solutions for our customers. By automating time-consuming tasks and enabling instant, scalable emissions calculations, we make it easier than ever for businesses to understand and reduce their carbon footprint. This innovative approach not only saves valuable time and resources but also drives greater efficiency and precision. We believe that using advanced machine learning is a critical step toward achieving meaningful carbon reductions at scale, empowering organizations to contribute to a more sustainable future.
Uses of Machine learning at Emitwise
Spend Product Classification Model
This model is a text classifier that classifies spend line items to a product from our product catalogue, so we can automatically assign emission factors and calculate emissions related to Scope 1, 2 and 3.
Key features:
- Built in-house
-
Exclusively trained on anonymised spend data, provided by customers and reviewed internally by Emitwise. This data is hosted in AWS, and adheres to Emitwise’s usual security and data protection guidelines
- Combined with in-house review by carbon accounting experts
Product Carbon Footprint (PCF) Product Classification Model
This model is a text classifier that classifies bill of material (BOM) line items to a product from our product catalogue, so we can automatically assign emission factors and calculate emissions related to materials.
Key features:
-
Uses OpenAI Completions for text classification. See how we use OpenAI here: https://academy.emitwise.com/emitwise-open-ai
-
Trained on publicly accessible bill of materials, product carbon footprint databases
-
Combined with in-house review by carbon accounting experts
- OpenAI is a sub-processor in this instance
Reduction Initiative Chatbot
This chatbot uses Emitwise’s database of publicly available reduction initiatives and case studies to give recommendations and answer the user’s questions related to them
Key features:
-
Uses OpenAI embeddings to embed user questions, retrieve relevant sections from cases studies related to the question, and OpenAI Completions to generate answers to the user’s questions. See how we use OpenAI here: https://academy.emitwise.com/emitwise-open-ai
- Trained on publicly available case studies and documented reduction initiatives. The content is sourced from reputable public databases and sustainability reports, curated to ensure the quality and reliability of the recommendations generated by the chatbot
- OpenAI is not a data sub-processor in this instance
Reduction Initiative Structurer
Developed in partnership with the WBCSD, this tool enables users to generate a case study following the WBCSD’s detailed structure for their case studies, from any file, user input or URL with information on the case study.
Key features:
-
Uses OpenAI Completions to extract content from a file and into dedicated fields. See how we use OpenAI here: https://academy.emitwise.com/emitwise-open-ai
- Trained on publicly accessible documents, URLs, or user-provided text files
-
Users can review and edit generated outputs before submission to WBCSD
Q&A
Q: What types of AI models are utilised?
A: Both proprietary and open-source models
Q:If the model is trained or fine-tuned, is customer data used?
A: Only anonymised customer data is used
Q: Is model performance evaluated?
A: Model evaluated continuously; models retrained regularly
Q: How frequently are AI models updated with security patches?
A: Immediately as needed
Q: How is data privacy and security enforced when using third-party AI models?
A: Contractual agreements and regular audits
Q: How do you ensure data quality, integrity, and security in AI systems?
A: Regular data audits and quality checks
Q: How is customer data persisted?
A: No data is persisted
Q: What security measures are in place to protect AI models, data, and infrastructure?
A: Multi-layered security with regular updates
Q: How frequently are your AI systems audited for compliance, security, and performance?
A: Annually