Over the past year, I've had some version of the same conversation with leaders across sustainability organizations. Some manage certification programs, others develop standards, oversee ESG initiatives, support healthy buildings, advance climate action, or help organizations measure and improve sustainability performance. Regardless of their mission, the question is usually the same:

"What can AI do for our organization?"

It's a good question but what I’ve noticed is that many of these organizations are jumping into AI tools and systems with very little scrutiny. And I get it - the pressure is real. Teams are stretched, resources are limited, and every vendor is promising huge fixes with the latest AI. But I firmly believe that organizations don’t need to learn responsible AI from scratch. They already understand governance. They know how to evaluate systems, establish policies, build trust, and protect the integrity of their work. AI should be held to those same standards.

That's why I believe responsible AI is about applying the principles your organization already uses every day to a new technology. Here is a playbook that could help as  you evaluate AI for your sustainability goals.  

1. Start with data governance

The underlying data that supports the AI is the most important thing you have going for you. Oftentimes, you’re working with customer data that is sensitive, you don’t want to inadvertently share customer data that isn’t meant to be public. When you introduce AI into these environments, the first questions aren't about model capability. They're about data:

• Where does your data go? If you're using a third-party AI platform, understand (contractually, not just in a sales conversation) whether your organization's data is retained, logged, or used to train future models.

• Do you have permission to use it this way? The organizations, members, customers, or partners who shared information with you may have agreed to one purpose, not necessarily AI processing. Make sure consent, contracts, and policies reflect how data is actually being used.

• Can you trace where AI outputs came from? When AI generates recommendations, summaries, or decisions, you should be able to identify the underlying information that produced those outputs. Traceability is just as important in AI as it is in any sustainability program.

Organizations will be far better served by treating data governance as an architectural decision rather than a policy document. Governance should be built into the platform itself through retention rules, access controls, audit logs, and clearly defined processing boundaries.

2. Protect against bias

Most practical AI today is some form of pattern recognition. These systems learn from historical data, and this data carries the fingerprints of everything that shaped past content, including things you may not want included. 

Imagine an organization that helps companies measure sustainability performance. Larger organizations typically submit cleaner, more complete data because they have dedicated sustainability teams. Smaller organizations often have fewer resources, different reporting formats, or less mature data collection processes.

If an AI model learns only from historical submissions, it may begin treating smaller organizations as inherently lower quality or higher risk because their documentation looks different, not because it’s worse. Over time, the AI could create additional problems and delays for exactly the organizations your mission is trying to support.

Here's a concrete example from operating a building certification platform across 100+ countries. Documentation styles vary enormously by region. A mechanical drawing prepared in one market looks different from the same drawing prepared in another. English fluency varies, local testing labs format reports differently, for example, and none of this has anything to do with whether a building actually meets a requirement.

But if your historical data shows that certification submissions from certain regions were resubmitted more often, perhaps because of language friction, or because reviewers were less familiar with local formats, a model trained on that history will learn to flag those regions more aggressively. It won't know it's learned geography, it will just be "more cautious" with certain submissions. And your program, whose mission is global, will have inadvertently created a system that makes certification harder for exactly the markets it's trying to grow in.

The fix is to test for this:

• Measure flag rates and false-positive rates across regions, languages, project types, and organization sizes before deployment and continuously after.

• Keep humans in the decision loop for anything consequential, and make sure the AI's suggestion doesn't become a rubber stamp. Automation bias is real.

• Give your program team, not just your tech team, visibility into these metrics. Bias in compliance decisions is a program integrity issue rather than a helpdesk ticket..

Bias in AI isn't just a technical issue, it's a governance issue that directly affects the integrity and credibility of your mission. 

3. Build trasparency in everywhere

AI will increasingly become part of how sustainability organizations operate. Whether it helps answer questions, review documentation, summarize reports, identify risks, or support decision-making, your reputation depends on how transparently those systems are used. 

Your brand is on the line, there is no other way to say this. If  you are using AI for major decisions and processes, the only way to ensure you can maintain your customer’s trust in this is through transparency. For sustainability organizations, you can build trust at three levels: 

To your stakeholders: Members, customers, partners, project teams, or program participants should understand when AI is involved and what role it plays. Most people are comfortable with AI assisting work. They become uncomfortable when they discover its involvement after the fact.

To your staff: Employees should understand why AI generated a recommendation, what information it considered, and how confident it is. They should be able to challenge or override recommendations easily. Those corrections become some of your most valuable learning data.

To your organization: Every significant recommendation or decision should have an audit trail that clearly distinguishes what AI suggested from what people ultimately decided. That accountability protects institutional knowledge, supports compliance, and strengthens trust over time.

We've learned some of these lessons firsthand. Early AI features were technically impressive, but they didn't always explain how they reached their conclusions. Some users ignored them. Others trusted them too much.

Both outcomes taught us the same lesson:
In sustainability organizations, explainability isn't a nice feature. It's a requirement. People need to understand why AI reached a conclusion before they can confidently use it to advance their mission.

4. The question your stakeholders will eventually ask

Here's an uncomfortable reality for sustainability organizations: AI systems consume energy, and your stakeholders know it.

If your organization helps others reduce emissions, improve environmental performance, disclose ESG data, advance healthy buildings, or promote sustainability best practices, someone will eventually ask a simple question:

"What about your AI?"

The good news is that responsible AI is often just good engineering.

Right-size the model to the task. Not every job requires the largest or most sophisticated model. Classifying documents, extracting information, checking completeness, or summarizing reports can often be handled by smaller, significantly more efficient models. In our experience, the biggest gains come from using the simplest model that reliably accomplishes the task and reserving larger models for work that truly requires advanced reasoning.

Batch and cache wherever possible. A surprising amount of AI compute is spent answering the same questions repeatedly. Eliminating unnecessary processing improves both efficiency and cost.

Measure what you use. You don't need perfect numbers, you simply need the same mindset your organization encourages in others: understand your resource consumption, look for opportunities to improve, and be transparent about your progress.

If I could leave sustainability leaders with one recommendation, it's this:
Document your AI governance framework before AI becomes embedded throughout your organization.

Define why your organization uses AI and just as importantly, where it should not be used.

Establish clear expectations around data governance, privacy, transparency, bias testing, human oversight, security, and operational efficiency. Determine who is responsible for reviewing those practices, how success will be measured, and how the framework will evolve over time as the technology changes.

The sustainability movement has earned credibility by demonstrating that good intentions alone are not enough. Progress requires measurable performance, accountability, and continuous improvement. AI deserves the same discipline.

The work sustainability organizations do is too important to rely on technology that isn't governed with the same care as everything else. Approaching AI in this way will reduce risk, and build greater trust with employees, partners, customers, funders, and the communities they serve and they'll help define what responsible AI looks like for the sustainability sector. The organizations that do will earn something far more valuable than operational efficiency. They'll earn trust.

And in sustainability, trust is still your most important asset.