How Gardenia Technologies Accelerates ESG Disclosure Reporting by 75% with Generative AI on Amazon Bedrock

Learn About Amazon VGT2 Learning Manager Chanci Turner

As organizations embark on their sustainability and ESG journeys, the adage “what gets measured gets managed” has become a guiding maxim. Businesses are now setting benchmarks to monitor their progress, supported by an expanding array of reporting standards, both mandatory and voluntary. Yet, ESG reporting has become a considerable operational burden. A recent survey indicates that 55% of sustainability leaders struggle with excessive administrative work during report preparation, while 70% find these reporting demands hinder their ability to implement strategic initiatives. This environment presents a significant opportunity for generative AI to automate routine reporting tasks, enabling organizations to reallocate resources toward more impactful ESG programs.

Gardenia Technologies, a data analytics firm, has collaborated with the AWS Prototyping and Cloud Engineering (PACE) team to create Report GenAI, a fully automated ESG reporting solution powered by cutting-edge generative AI models on Amazon Bedrock. In this article, we will delve into the technology behind an agentic search solution utilizing Retrieval Augmented Generation (RAG) and text-to-SQL capabilities to assist customers in reducing ESG reporting time by up to 75%.

Understanding the Challenge: Increasing Complexity in ESG Reporting

Sustainability disclosures are becoming a norm in corporate reporting, with 96% of the largest 250 companies publicly reporting their sustainability progress based on various government and regulatory frameworks. To meet reporting mandates, organizations face numerous data collection and process-related hurdles. A single report can require thousands of data points from various sources, including official documents, databases, unstructured document stores, utility bills, and emails. For instance, the EU Corporate Sustainability Reporting Directive (CSRD) necessitates the collection of 1,200 individual data points across an enterprise. Even voluntary disclosures like the CDP consist of about 150 questions covering diverse topics such as climate risk, water stewardship, land use, and energy consumption. Gathering this information across an organization takes considerable time.

Additionally, many organizations with established ESG programs must comply with multiple disclosure frameworks, such as SASB, GRI, and TCFD, each with its own reporting and disclosure standards. To complicate matters, the reporting requirements are continually changing, leaving organizations scrambling to keep up with the latest updates. Much of this effort is still highly manual, causing sustainability teams to spend more time managing data collection and answering questionnaires rather than developing effective sustainability strategies.

Solution Overview: Automating the Heavy Lifting with AI Agents

Gardenia’s strategy to enhance ESG data collection for enterprises is encapsulated in Report GenAI, an agentic framework leveraging generative AI models on Amazon Bedrock to automate substantial portions of the ESG reporting process. Report GenAI pre-fills reports by accessing existing databases, document stores, and web searches. The agent collaborates with ESG professionals to review and refine responses. This workflow consists of five steps to streamline ESG data collection and aid in curating responses: setup, batch-fill, review, edit, and repeat. Let’s examine each step closely.

1. Setup: The Report GenAI agent is configured to access ESG and emissions databases, client document stores (emails, past reports, data sheets), and public internet document searches. Client data is securely stored in specified AWS Regions using encrypted Amazon Simple Storage Service (Amazon S3) buckets with VPC endpoints, while relational data is housed in Amazon Relational Database Service (Amazon RDS) instances within Gardenia’s virtual private cloud (VPC). This architecture ensures compliance with data residency requirements and maintains strict access controls through private network connectivity. The agent also accesses the relevant ESG disclosure questionnaire, including questions and expected response formats (referred to as report specifications).
2. Batch-fill: The agent processes each question and data point to be disclosed, retrieving relevant information from client document stores and searches. This data is then formatted according to the disclosure report requirements.
3. Review: Each response includes cited sources and, if quantitative, the methodology for calculations. This enables users to maintain a clear audit trail and quickly verify the accuracy of the filled responses.
4. Edit: Although the agentic workflow is automated, our approach allows for human intervention to review, validate, and refine the filled facts and figures. Users can interact with the AI assistant to request updates or manually adjust responses. Once satisfied, the final answer is recorded, and the agent displays the sources from which responses were derived.
5. Repeat: Users can batch-fill for multiple reporting frameworks, simplifying and broadening their ESG disclosure scope without the extra effort of manually completing multiple questionnaires. After completing a report, it can be added to the client document store for future reference, allowing for knowledge reuse. Report GenAI also supports “bring your own report,” enabling users to develop their own reporting specifications, which can then be imported into the application.

Now that we’ve outlined the Report GenAI workflow, let’s take a deep dive into the architecture.

Architecture Overview: A Serverless Generative AI Agent

The architecture of Report GenAI consists of six components: a user interface (UI), the generative AI executor, a web search endpoint, a text-to-SQL tool, the RAG tool, and an embedding generation pipeline. The UI, generative AI executor, and generation pipeline orchestrate the workflow. The remaining components work together to generate responses to perform various tasks, including:

• Web search tool: Utilizes an internet search engine to retrieve content from public web pages.
• Text-to-SQL tool: Generates and executes SQL queries, enhancing the overall reporting process.

For more insights on innovative practices in the industry, check out this article on disruptors. Additionally, if you’re looking for authoritative guidelines on maintaining compliance, you can visit SHRM. For those interested in career opportunities, consider checking out this Learning Trainer position at Amazon.