Sustainability information is unlike traditional financial information in one important way: it often comes from everywhere.

Greenhouse gas emissions data may sit with operations, facilities, procurement, fleet management, travel providers, landlords, suppliers, and finance. Workforce data may come from HR. Waste data may come from facilities or third-party service providers. Climate risk information may come from enterprise risk management, strategy, legal, finance, and business units. Community, nature, supply chain, and governance metrics may each have their own data owners, systems, assumptions, spreadsheets, and interpretations.

This is what makes sustainability reporting both powerful and vulnerable. When done well, it can provide valuable insight into business performance, risk, resilience, efficiency, and long-term value creation. When done poorly, it can become a patchwork of numbers collected for disclosure purposes, without enough clarity on ownership, methodology, review, or control.

The rapid rise of artificial intelligence has added a new layer to this conversation. AI has enormous potential to improve how organizations collect, analyze, and use sustainability information. It can help identify anomalies, summarize large volumes of supplier data, support scenario analysis, streamline evidence gathering, draft narrative disclosures, and surface insights that might otherwise be missed. For sustainability teams that are often under-resourced and managing growing reporting demands, the appeal is obvious.

But there is also a risk: moving too fast.

AI should not be used to automate a weak process. It should be used to strengthen a well-governed one.

Before organizations rush to apply AI to sustainability metrics, they need to ask a more basic question: Do we trust the underlying data, process, and controls?

If the answer is no, AI may simply help organizations produce questionable information faster.

A sustainability metric is only as reliable as the system behind it. Consider something as seemingly straightforward as energy consumption. On the surface, the organization may believe it has a clear metric. But where does the data come from? Utility bills? Landlord estimates? Meter readings? Manual entries? Are leased assets included? Are estimates documented? Are units standardized? Who reviews unusual fluctuations? Has the methodology changed from the prior year? Are there controls to ensure completeness across all sites?

If these questions are not answered before automation is introduced, AI may create a false sense of confidence. The output may look polished and sophisticated, but the underlying information may still be incomplete, inconsistent, or unsupported.

This is particularly important because sustainability information is increasingly being used for more than voluntary reporting. It is feeding investor communications, regulatory disclosures, executive decision-making, capital allocation, financing discussions, risk management, and assurance processes. As expectations rise, sustainability data needs to move closer to the discipline historically applied to financial information.

That does not mean sustainability reporting needs to become overly bureaucratic. It means organizations need fit-for-purpose governance.

At a minimum, organizations should be clear on who owns each metric, who prepares it, who reviews it, and who approves it (see previous blog post on this topic). They should document methodologies, define boundaries, maintain evidence, and track changes in assumptions (see blog post on this topic). They should understand where manual intervention occurs and where errors are most likely to arise. They should also identify which metrics are most material to the business, stakeholders, and regulatory environment, rather than trying to apply the same level of control to every data point at once.

This is where AI can become far more effective.

Once the governance foundation is in place, AI can help organizations move from reactive reporting to better insight. It can flag outliers in emissions data, identify missing supplier responses, compare current performance against prior periods, summarize large volumes of climate-related information, support internal dashboards, and help teams understand where data quality is improving or deteriorating. It can also reduce the administrative burden of sustainability reporting, freeing teams to focus on analysis, strategy, and decision-making.

For example, AI could help a company identify that electricity use at one site has increased significantly compared to prior periods. But the value of that insight depends on whether the organization has confidence that the site was included consistently, that the source data is complete, that the units were converted correctly, and that someone is accountable for investigating the variance. Without those controls, the AI-generated insight may create more confusion than value.

The same applies to supplier emissions, waste diversion, water consumption, safety metrics, diversity data, or climate risk inputs. AI can help process and interpret information, but it cannot replace the need for clear accountability, sound methodology, and human judgment.

There is also a governance risk in relying too heavily on AI-generated narrative. Sustainability disclosures require context, balance, and accuracy. AI may help draft or structure disclosure language, but organizations still need controls over what is being said, whether claims are supported by evidence, and whether the disclosure fairly reflects performance. This is especially important in an environment where greenwashing risk, litigation risk, and regulatory scrutiny are increasing.

The opportunity is not to choose between AI and controls. The opportunity is to bring them together.

Organizations that will benefit most from AI are not necessarily those that adopt it the fastest. They are the ones that understand their sustainability data landscape, build strong governance around their most important metrics, and then use AI intentionally to improve efficiency, consistency, and insight.

1) A practical approach is to start with the metrics that matter most rather than trying to automate everything at once. Identify the sustainability information that is material to the organization’s strategy, risk profile, regulatory obligations, and stakeholder expectations.

2) Before introducing AI, the organization should map how the data currently moves through the business. For each metrics, identify the following:

This step often reveals that sustainability data is more fragmented than expected. For example, energy data may come from facilities, leased locations, utility bills, landlord estimates, and regional teams all using slightly different formats or assumptions.

This is exactly why AI should not be added too early. If the process is unclear, AI may automate inconsistency.

3) Once the data journey is mapped, assess where the process is vulnerable.

Common issues include:

• Missing sites, entities, vehicles, or suppliers
• Inconsistent units of measure
• Manual spreadsheet manipulation
• Unclear assumptions or estimates
• Lack of version control
• No formal review process
• Methodology changes that are not documented
• Data owners who do not understand how their inputs affect reporting

This does not need to be overly complex. A simple risk assessment can classify each metric as low, medium, or high risk based on complexity, judgment, manual effort, materiality, and external scrutiny. The goal is to understand where could this metric be incomplete, inaccurate, inconsistent, or unsupported?

4) Once these issues are identified, clear governance and internal control processes need to be in place if they weren’t already in step 2. Accountability around the metrics makes AI adoption safer and more reliable.

Finance, sustainability, operations, risk, IT, and internal audit can each play a role here and each metric should have:

• A named data owner
• A documented methodology
• Defined reporting boundaries
• Clear calculation logic
• Evidence requirements
• Review and approval steps
• Version control
• Change management for methodology updates
• Escalation procedures for unusual results

My previous blog posts provide many further details around each of the points above.

5) Once the organization is comfortable with the above, introduce AI in targeted ways. For example to summarize supplier responses, draft first-pass narrative disclosures, support scenario and sensitivity analysis. Human eyes are still essential and Ai should never be the final decision point. The AI should also be pilot-tested prior to going large-scale. For example if it is analyzing data and detecting outliers and anomalies, have it documented as to how mistakes are dealt with, what data the AI has access to and what controls there are over the AI itself ( ie: who reviews the information etc..). The same internal controls and governance practices we discussed above and in previous blogs apply to the AI itself.

The danger is not AI itself. The danger is using AI to accelerate processes that were never properly designed, governed, or controlled in the first place. Used properly, AI becomes an enabler of better sustainability management, not a shortcut around it.

As sustainability reporting matures, organizations need to resist the temptation to move straight to automation before addressing the fundamentals. Reliable sustainability information requires the same disciplines that support good business information: ownership, consistency, documentation, review, accountability, and governance.

AI can help organizations do more with sustainability information. But first, organizations need to make sure the information is worth relying on.