Second place AIM Hackathon: Trustpilot for ESG – Detecting Greenwashing with AI 🌱

 Authors: Mark Drozdov, Artur Sogomonyan and Ilya Belov

Why greenwashing?

Greenwashing is a widespread problem. Companies often claim to be “sustainable” or “eco-friendly” without proper evidence. The developed solution aims to automatically detect greenwashing by analyzing ESG reports and assigning a trustworthiness score.

Key Types of Greenwashing To Detect:

• Z1 Hidden Alternative Costs (Compromises)

• Z2 Lack of Evidence in Communicating Marketing Activities

• Z3 Lack of Accuracy

• Z4 Lack of Transmission into Real Product Features

• Z5 Lesser Evil

  The team focused on these five because they’re easily detectable in text. They found clear examples of each in H&M’s 2021 Sustainability Report, previously flagged by the Changing Markets Foundation for greenwashing.
   

How RAG Works:

1. Document Upload: Users upload an ESG report, and the system extracts and cleans the text.

2. Retrieval: RAG pulls relevant examples from our dataset, which contains pre-labeled greenwashing examples.

3. Generation & Detection: The AI compares the new report to the retrieved examples and categorizes potential greenwashing.

4. Highlighting & Scoring: The system highlights greenwashing sections and assigns a trustworthiness score.

5. Continuous Improvement: With every new report uploaded, the system learns and improves.