Your AI advantage will come from governance, not performance

AI agents are rapidly moving beyond simple assistance. They now update records, initiate workflows, recommend suppliers, flag operational issues and carry out business processes. That’s why organizations need the confidence to understand an automated workflow, review the data that informed it and assign responsibility if the system gets something wrong. “For the mission-critical processes of our customers, ‘almost right’ just isn’t good enough,” said Christian Klein, SAP CEO, about the shift to an autonomous enterprise.    

Therefore, the next competitive advantage in AI may be less about access to more powerful models and more about whether a company can deploy it with confidence. That requires credible safeguards around privacy, security, bias, human rights, transparency and environmental impact. Strong governance enables a company to answer those questions before a customer or regulator asks. It sets expectations before AI systems become embedded in operations.

The gap between policy and practice

Many companies have adopted responsible AI principles. Far fewer have translated those principles into controls that influence real products, procurement and operating decisions.

S&P Global’s 2025 Corporate Sustainability Assessment found that 36 percent of responding companies had either a dedicated AI policy or an AI policy integrated into wider corporate governance. While a policy does not establish effective oversight on its own, it can define who owns risk, which systems are covered, what teams must assess an AI use-case before deployment and when further review is required.

Even when policies exist, environmental issues receive limited attention. BSI research found that 27 percent of surveyed organizations consider AI’s footprint in sustainability strategy and reporting. That gap is becoming harder to ignore, as data-intensive workloads increase demand for electricity, data center capacity, cooling and hardware. 

For sustainability leaders, this creates a practical governance challenge. If environmental concerns enter only after technology decisions have been made, companies are left with efficiency optimizations alone – such as smaller prompts, better caching, or right-sized models. These matter, but they are only part of the picture. The higher-leverage decisions concern whether AI is needed for a given task – and what type of AI. Those questions can only be asked before the use case is scoped.

Sustainability should part of the initial question: Should this system be built? What resources will it require? What safeguards are needed before it becomes embedded in a business process?

Turning principles into practice

Credible and effective AI governance requires defined decision points that product, technology, legal, risk and sustainability teams use together. The useful lesson for the broader market is the importance of a workflow that converts stated principles into accountable decisions.

A practical operating model includes:

• A required impact assessment before a use case moves into production. This should distinguish between standard systems and high-risk systems.
• A cross-functional review process with authority to require mitigation, redesign, a delayed launch or rejection. This should pair with human oversight requirements that clarify when an employee approves, overrides or stops an automated action.
• Logs and documentation that enable teams to investigate actions, data inputs and outcomes. These should be supported by environmental criteria including energy use, emissions, water use, hardware demand and expected growth in workload.
• Public reporting that shows how governance decisions affect deployment and operational performance.

Under SAP’s Global AI Ethics Policy, for example, every AI use case must undergo a rigorous evaluation before deployment, based on SAP’s 10 AI Ethics principles that are routed in the UNESCO recommendations on the Ethics of AI. Evaluated use cases are classified according to their risk, and the AI Ethics Office helps product teams put the necessary safeguards in place.

Depending on the risks, some use cases undergo several review steps. They may be escalated to SAP’s Global AI Ethics Steering Committee or even to the Executive Board for a final decision. In 2025, SAP assessed 860 AI use cases; 44 percent of them were classified as high-risk and referred to an in-depth assessment, according to the Integrated Report. Where the review found no ethically acceptable tradeoffs or ways to reduce the risk, the project could not proceed in its original form.

With this, SAP’s policy incorporates sustainability, human oversight, transparency, privacy and accountability principles. This alignment with UNESCO’s recommendations on the Ethics of Artificial Intelligence has earned SAP external recognition. However, such an example is rare. For instance, among 2,972 companies, only 12.4 percent maintain an explicit policy ensuring human oversight on AI systems, according to benchmark research by the Thomson Reuters Foundation and UNESCO. SAP is one of them.

Environmental accountability is next

Environmental reviews belong in the same decision processes as performance, cost, privacy and security. They should begin parallel with the use-case design, when teams still have a meaningful chance to embed the needed considerations.

Companies can ask these questions to their teams:

• Can the workload be designed more efficiently by reducing unnecessary data processing and inference volume, or by selecting a smaller model that meets business needs?
• What energy, emissions, water and hardware demands will arise from the use case?      Can cloud, and model providers supply sufficient data to measure and report those impacts accurately?
• Are the expected social or operational benefits proportionate to the use case’s resource demand? How do governance structures ensure that AI scaling decisions remain aligned with the company’s climate commitments, and who is responsible for monitoring and reporting on that balance over time?

The answers will vary by use case. A high-volume service application, a finance workflow and a tool that improves industrial energy management require different assessment criteria. Explicitly evaluating the tradeoffs between risks and benefits gives companies more credibility than declaring AI innovations sustainable from the get-go.

Actions for leaders

The industry does not need another vague commitment to responsible AI. It needs governance that holds up under real deployment pressure.

Business, technology and sustainability leaders can begin by identifying where AI is already affecting material operations, customers, employees and resource use. They can then establish an assessment process, give sustainability teams formal decision rights and define escalation criteria. They should also request consistent environmental disclosure from AI and cloud providers.

Leaders should also be candid about where measurement remains incomplete. Environmental data across AI value chains can be limited, supplier methodologies differ, and operational impacts may change quickly as usage grows. But incomplete data is not a reason to delay action: cost-aligned optimisations such as reducing inference volume and right-sizing models already deliver environmental benefits while broader measurement catches up. A transparent account of those constraints supports better governance and clearer expectations for investors, customers, employees and companies alike.

In the agentic era, trust is your organization’s license to operate. Governance is how you earn it.