Sustainability

Beyond the Hype: How Sustainability Leaders are Actually Deploying AI

For sustainability professionals, the discourse surrounding Artificial Intelligence often drifts into the realm of the abstract—a flurry of high-level promises about "optimizing the planet" that rarely translates into daily workflows. The gap between theoretical potential and practical utility has left many in the ESG (Environmental, Social, and Governance) sector feeling paralyzed by the complexity of implementation.

Seeking to cut through this noise, I recently convened a roundtable of sustainability leaders from the Trellis Network. The mandate was simple: skip the theory and provide a granular account of what they have actually built, how they built it, and the hurdles they encountered along the way.

The findings were revealing. Rather than relying on bespoke, multi-million dollar software packages, most participants were leveraging existing enterprise tools—Microsoft Copilot, Google’s NotebookLM, and OpenAI’s ChatGPT. These are solutions already licensed by their employers, requiring no additional procurement budgets or specialized data science teams. This shift signals a democratization of AI in the sustainability space, proving that the most effective tools are often already at our fingertips.

The Practical Mechanics: Two Case Studies

To understand how these tools operate in the wild, two members of the Trellis Network shared their experiences in deploying AI agents to solve specific, high-friction operational challenges.

Griffith Foods: Empowering the Sales Frontline

Griffith Foods, a global leader in seasonings, sauces, and dressings for the food industry, faced a recurring challenge: their business development and sales teams frequently struggled to articulate the company’s complex sustainability and nutrition goals in a way that resonated with clients.

Madeline Patterson Smith, the company’s global director of sustainability impact reporting, sought to bridge this gap. She built a Copilot-based agent designed to synthesize Griffith Foods’ "2030 Aspirations"—a rigorous action plan covering nutritional health and environmental impact—into accessible, client-facing narratives.

The Development Process:
Patterson Smith built the agent solo, working incrementally between her primary duties during annual reporting cycles. To ensure corporate compliance, she implemented strict guardrails:

  • Controlled Data Silos: The agent only accesses specific SharePoint folders curated by Patterson Smith, rather than searching the open internet.
  • The "Honesty" Protocol: She programmed the agent to explicitly state "I do not know" rather than hallucinating or guessing when an answer fell outside its provided source documents.

The Hurdles:
The journey was not without technical friction. Initially, Patterson Smith attempted to integrate live Power BI dashboards. The agent, however, struggled to discern which data points were current, often pulling outdated figures. She pivoted to a "snapshot" method, uploading static, month-end data files, which solved the confusion. Additionally, the agent demonstrated a "preference" for information stored in standalone Word documents over data buried within complex PDFs or slide decks, highlighting the importance of document structure when training AI agents.

TE Connectivity: Mastering Cross-Border Data Synthesis

At TE Connectivity, a manufacturer of essential electrical components, the hurdle was organizational fragmentation. Preparing for the company’s next corporate responsibility report required synthesizing dozens of disparate documents—website analytics, PR reports, and performance data—from China, Japan, and the U.S.

Brendan Stelmach, the global corporate responsibility manager, realized that manual synthesis would be a "nightmare." Instead, he leveraged Copilot to act as an analytical engine.

The Analytical Workflow:
Stelmach uploaded approximately 36 files into a single repository. He then prompted the AI to perform two specific analyses: extracting key engagement metrics and identifying regional content preferences. The task, which would have taken a team weeks, was completed in a single afternoon. A pivotal discovery—that 90 percent of sustainability page traffic originated from desktop users—confirmed that their audience was comprised of professional stakeholders rather than casual browsers, fundamentally shifting their reporting strategy.

Stelmach offered a critical insight for those looking to replicate his success: model selection matters. He found that routing complex analytical tasks to Anthropic’s Claude models yielded significantly higher quality results than using the default model, proving that "prompt engineering" is secondary to choosing the right architectural engine.

Chronology of the AI Shift in Sustainability

The rapid adoption of these tools follows a distinct, three-phase timeline that has accelerated over the last 18 months:

  1. The Exploratory Phase (Q1–Q2 2023): Teams began using ChatGPT for basic drafting and brainstorming. Early adopters tested the boundaries of data privacy, leading to the establishment of "private" corporate instances.
  2. The Tooling Phase (Q3–Q4 2023): Large enterprises integrated Copilot and Gemini into their Microsoft 365 and Google Workspace environments. Sustainability teams began experimenting with "RAG" (Retrieval-Augmented Generation) by pointing these tools at internal SharePoint repositories.
  3. The Implementation Phase (2024–Present): Sustainability leaders moved from "chatting" with AI to building specialized agents that perform specific, repeatable tasks like greenwashing detection, supplier data collection, and report synthesis.

Supporting Data: The Impact of AI on Efficiency

While AI is often associated with "creative" tasks, its most profound impact in the sustainability sector is administrative efficiency.

  • Supplier Engagement: Western Digital’s pilot with the startup Sluicebox demonstrated that AI agents could increase the collection of primary carbon data from suppliers from 30 percent to 90 percent, while reducing the timeline for data gathering from six months to four weeks.
  • Quality Control: Google’s internal "claims verifier" allows for automated flagging of vague terminology in marketing, ensuring that sustainability communications meet both internal standards and external regulatory requirements (such as EU greenwashing directives).
  • Resource Management: For companies like Eastman, deploying a "Green Claims" chatbot allows for a proactive rather than reactive approach to compliance, screening packaging copy against FTC guidelines before it hits the production floor.

Official Perspectives and Ethical Guardrails

The consensus among these leaders is that AI is an assistant, not a replacement for human judgment. As Patterson Smith noted, "It is much easier than I thought, but it does take a lot of iterating with teams."

The primary concern among sustainability officers remains the "black box" of algorithmic bias. The cautionary tale of an agent built to summarize biodiversity guidance—which began inventing resources and framing corporate negligence as "industry-leading practice"—serves as a stark warning. The resolution for such issues, according to industry experts, is the implementation of an "intake layer." By forcing the agent to categorize the user’s sector, geography, and risk appetite before providing an answer, the AI is effectively "caged" within a relevant, fact-based context.

Broader Implications for the Profession

The shift toward "DIY AI" has profound implications for the future of sustainability roles.

1. From Data Collector to Data Architect:
The traditional sustainability role—often defined by the drudgery of data collection and manual report formatting—is being automated. In its place, the role of the "Sustainability Architect" is emerging: someone who designs the AI workflows, manages the document sources, and ensures the accuracy of the outputs.

2. The End of "Report Fatigue":
As TE Connectivity’s experience demonstrates, the ability to synthesize data from multiple global regions into a single, cohesive narrative allows companies to produce more frequent, more specific updates rather than waiting for massive, annual disclosures.

3. Internal Governance as a Competitive Advantage:
As more companies adopt AI, the "cleanliness" of their internal data becomes a competitive differentiator. Companies with well-organized, digitized, and tagged data will be able to deploy AI agents that work flawlessly, while those with messy, siloed, or analog documentation will continue to struggle with "hallucinating" or inaccurate agents.

4. Addressing the "Everyone is Doing It" Problem:
As Brendan Stelmach observed, the accessibility of these tools can lead to internal redundancy. Companies must now implement internal governance to ensure that sustainability, marketing, and IT teams are not building identical agents in silos, which leads to fragmented data handling and wasted resources.

Conclusion

The era of waiting for specialized, expensive ESG software is over. The current reality is that sustainability teams possess the agency to build their own tools using the infrastructure they already own. By focusing on narrow, high-value tasks—such as starting sales conversations, synthesizing global reports, or screening marketing claims—sustainability professionals are successfully moving past the theoretical hype.

The most successful implementations, as seen at Griffith Foods and TE Connectivity, share a common thread: they do not try to make the AI "know everything." Instead, they treat the AI as a diligent, albeit literal-minded, clerk that operates only within the boundaries of the verified, company-controlled documents provided to it. In doing so, these leaders are proving that the most sustainable path forward is one where human expertise is augmented, not replaced, by the machine.