5 ways companies can promote more sustainable AI

Artificial intelligence (AI) is powering one of the biggest transformations in history, but powering AI is vast amounts of energy—an estimated 85.4 terawatt-hours of electricity annually by 2027, or more than most small countries use in a single year.

As the demand for computing grows and accelerates, so too does the potential environmental cost. According to a recent MIT study, training a single AI model can emit as much carbon as five cars in their lifetimes. Put simply, higher energy consumption means higher carbon emissions, a slippery slope as we struggle to get rising global temperatures under control.

Despite the risks, there are enormous opportunities. AI indeed has the potential to vastly increase our carbon footprint, but finding more sustainable approaches to its development can help companies use it more responsibly—and even help to combat emissions in the long run.

From optimizing our energy consumption to improving climate models and detecting methane leaks, AI has a critical role to play in climate change mitigation and adaptation. Particularly as the energy used to train and deploy its models increasingly comes from sustainable sources. A BCG study found that AI can actually reduce an organization’s carbon footprint by 5 to 10 percent, or by up to 5.3 gigatons of carbon dioxide equivalent (CO2e) if scaled globally. That’s roughly the emissions of India and Japan combined.

Small steps, big impact

At a business level, AI can be used to automate and optimize operations, which helps to reduce carbon emissions at scale. Since launching our “suggested macros” feature in October 2021, Zendesk has saved an estimated 18.6 million minutes of agent time for our customers. This in turn has helped to reduce over 600 tonnes of CO2e from customer experience (CX) activities—roughly equivalent to 133 gasoline-powered vehicles driven for one year. And that’s just the impact of one feature.

We believe that AI can be a powerful tool in the push to accelerate climate action, but only if used and developed responsibly. Here are the steps companies can take to promote more sustainable AI:

1. Boost awareness

   Raising stakeholder awareness is a critical first step to driving action. By implementing responsible AI policies and educating employees, companies can promote a more sustainable culture around AI training and implementation.

   At Zendesk, environmental sustainability is a key consideration of our Business Code of Conduct and our guidelines for responsible AI usage.

2. Call for greater supplier transparency

   Companies should call on their cloud vendors and AI service providers to be more transparent about their own AI policies, and push for more sustainable practices where possible. This sends a strong signal to suppliers that more accurate data around the true environmental impact of AI services is needed.

3. Promote energy efficiency

   Using energy-efficient IT hardware, such as servers and laptops with a high Energy Star rating, can play a major role in minimizing power usage of AI computation and in reducing companies’ carbon footprint. It can also greatly reduce companies’ energy expenses.

4. Use renewable energy

   When possible, power AI computation with renewable energy, such as solar and wind. These sources produce energy without depleting natural resources and with minimal greenhouse gas emissions.

5. Apply sustainable engineering practices

   AI engineers can help to improve energy efficiency and sustainability across the entire development lifecycle. At Zendesk, we have used the following techniques to reduce the carbon footprint of our AI features:

• Use existing models and trained datasets whenever possible: Developing, training, and deploying new models requires additional energy and risk contributing to carbon emissions.

• Rightsize and simplify AI models: Large language models (LLMs) are not always the best approach to solving problems. Consider using smaller, lighter models to reduce computation power, while also accomplishing business outcomes.

• Use domain-specific models tailored to our business needs: This can lead to similar outputs with much better efficiency. For instance, Zendesk AI uses models that are trained specifically for the CX industry.

• Only use AI when we truly need it: Think about whether AI is the best tool to solve the problem, or whether other technologies (with lower computing and energy requirements) can be used to achieve business needs.

Our shared role

AI will transform much of how we live and work. That much is certain. But how it’s used and to what end have yet to be determined.

We all have a role to play in ensuring that AI can contribute to a more sustainable, low-carbon future—and what we do now matters. Zendesk is committed to measuring and reducing the environmental impact of our AI products, and as we do so, we will continue to share our learnings with our communities and customers.