AI and Sustainability: What It Actually Does Across Carbon Accounting, ESG and LCA

Why AI Feels Like Pressure, Not Progress

I’ve realised over the past few months that every time AI comes up in a conversation, it doesn’t really simplify things - it adds a different kind of pressure. Someone mentions using AI for something, and the expectation shifts almost immediately. You’re supposed to understand it, apply it, and somehow get better outcomes because of it, even when the details aren’t entirely clear.

It also doesn’t stay contained. It shows up across tools you’re already using, in new platforms being introduced, and in conversations where “AI-powered” gets attached to almost everything without much clarity on what that actually means in practice.

In sustainability, that gap feels wider. Most teams are still working through the basics - getting consistent data, aligning on methodologies, responding to different frameworks - and now AI is being layered on top as if it’s the missing piece that will make all of this easier.

The problem is that it’s not always clear what it is actually solving, and what it might be quietly making more complicated.

What AI Is Actually Doing Behind the Scenes

Once you start working with these tools more closely, it becomes clear that “AI” in this context is not one single capability, even though it is often presented that way. Most of what is being described as AI is handling specific, well-defined tasks in the background. It might involve structuring messy datasets, mapping inputs to reference points, or flagging values that do not align with expected patterns.

This kind of support is useful because it reduces the time it takes to get to a working dataset. Tasks that would otherwise take weeks can move faster, and some of the more repetitive parts of the work become easier to manage. At the same time, the limits become visible quite quickly. These systems do not correct poor-quality input data, they do not make methodological decisions, and they cannot reliably account for regional or sector-specific gaps if that data does not exist in the first place.

In most cases, the quality of the output still depends heavily on the databases being used and the people reviewing the results. That becomes particularly important in sustainability work, where relatively small assumptions can change outcomes in ways that are not always immediately obvious. It is entirely possible to produce outputs that look structured and complete, but are directionally off because something upstream was not fully understood or verified.

Where AI Starts to Help

Where AI starts to make sense is not in replacing the work, but in reducing the friction around it. Most sustainability teams are not struggling because they do not know what needs to be done. The difficulty is that a large part of the effort goes into getting data and processes into a state where they can actually be used.

Making messy data usable

A common issue is how unstructured the starting point tends to be. Sustainability data rarely arrives in a clean, consistent format. It is pulled from different systems, labelled in different ways, and often missing context. Working with it can feel less like analysing a dataset and more like organising disconnected pieces before anything meaningful can begin - similar to trying to work from a pile of unsorted documents where nothing is labelled properly.

AI helps here by bringing a level of structure to that data. It can sort, label, and group inputs, so that teams are not starting from scratch each time. It does not make the data complete or correct on its own, but it creates a usable baseline that allows the rest of the work to move forward.

Reducing repeated work across requests

Another pattern is the amount of repetition in sustainability workflows. The same underlying information is used across internal reporting, customer requests, investor questionnaires, and regulatory disclosures, but each of these requires it in a slightly different format. This often feels like rewriting the same answer again and again depending on who is asking.

AI can reduce that effort by making it easier to adapt and reorganise information across formats, much like having a draft that can be reshaped instead of rewritten from scratch each time. It does not eliminate the need for review, but it reduces the amount of manual restructuring that happens with every new request.

Handling scale without losing control

The third area is scale. Processes that are manageable at a smaller level begin to break when they expand to cover hundreds of suppliers, products, or data points. At that point, maintaining consistency and traceability through manual effort becomes difficult, similar to trying to track everything manually once the volume grows beyond what you can realistically keep in your head or in a spreadsheet.

AI helps by handling larger volumes of data, generating initial outputs, and highlighting areas that need attention. It allows teams to operate at a scale that would otherwise slow everything down significantly, while still leaving space for validation and refinement where it matters.

Across all of these cases, the role of AI is relatively consistent. It helps teams get to a usable starting point faster. What it does not do is remove the need to understand, validate, and stand behind the results.

How This Shows Up in Sustainability Work

When you move from general use cases to actual sustainability workflows, the role of AI becomes much more concrete. It is not evenly distributed across the work. It shows up very clearly in specific areas, and much less in others. Understanding where that difference sits is what makes it useful rather than confusing.

Carbon accounting

In carbon accounting, the most immediate impact of AI is in emissions mapping and classification. A large part of the effort typically goes into matching business activity data - procurement records, invoices, logistics entries - to the right emissions factors. Doing this manually is slow and often inconsistent.

AI changes this by enabling automatic emissions factor matching at scale, where inputs are mapped to the closest relevant factors with a level of consistency that is difficult to maintain manually. This does not remove the need for review, but it significantly reduces the time required to get to a first usable inventory, especially in Scope 3 where the volume and variability of data is highest.

ESG reporting

In ESG reporting, the challenge is not just data collection, but how that data is used across multiple frameworks. The same underlying information needs to be adapted for different formats - investor questionnaires, regulatory disclosures, customer requests - each with its own structure and expectations.

AI helps by enabling faster mapping across reporting formats and by identifying peer benchmarks and best practices, allowing teams to respond more efficiently while maintaining consistency.

A simple example is the overlap between CSRD and GRI disclosures. Both frameworks require reporting on emissions and governance structures, but the way they ask for that information differs in structure and level of detail. Without any support, teams often end up reworking the same dataset to fit each format. AI can help by recognising these overlaps and adapting the same underlying data into different disclosure formats, reducing the need to rebuild responses each time.

Product-level work (LCA / PCF)

At the product level, the challenge shifts to scale and repeatability. Life cycle assessments and product carbon footprints require multiple inputs, assumptions, and process steps, which can feel difficult to structure, especially when done for the first time.

AI helps here by breaking down complex processes into structured workflows and generating ready-made templates or first-cut models. In industries like chemicals, where production stages follow relatively consistent patterns, this becomes particularly useful. Instead of starting from scratch, teams can begin with a baseline model and then adapt it with product-specific data.

This makes product-level carbon analysis more manageable and repeatable, rather than a one-off exercise each time.

Across all three areas, the role of AI becomes clearer when you look at it this way. It is most valuable in getting teams to a structured starting point faster. What still matters - and often matters more - is how that output is reviewed, adapted, and ultimately trusted.

Where Things Still Break

Even with all of this, there are parts of the workflow where AI does not just fall short, but can create a false sense of confidence if it is not used carefully. The outputs can look structured and complete, which makes it easy to assume that the underlying work is equally robust, even when it is not.

The first issue is input quality. AI can organise, map, and process what it is given, but it does not correct flawed inputs. If a supplier provides incomplete data, or if a dataset is built on assumptions that are not clearly defined, the system will still produce an output that appears consistent. The gap only becomes visible when the numbers are examined more closely, often at a stage where changes are harder to make.

The second issue is methodology and interpretation. Sustainability calculations are not purely mechanical. Decisions around system boundaries, allocation methods, or how emissions are distributed across products require context and judgement. These are the areas that typically come under scrutiny during audits or assurance reviews, and they are not decisions that can be reliably automated.

The third issue is regional and sector-specific accuracy. A large proportion of emissions factor databases and reference datasets are built on European and North American industrial data. When organisations operate in regions like India, the UAE, or Saudi Arabia, the closest available factors are often proxies rather than precise matches. AI can still generate outputs using those proxies, but it does not always make that limitation obvious, which can result in outputs that look complete while being only partially representative.

Across all three areas, the pattern is consistent. AI does not fail visibly. It fails quietly, by producing outputs that appear reliable but are built on assumptions that have not been fully examined.

That is why the more useful question is not whether a system uses AI, but where it is being relied on, and where human judgement still sits.

Why This Is Becoming Harder to Ignore

A few years ago, slower and less connected sustainability workflows were frustrating, but still manageable. Teams could work with partial data, take more time to refine outputs, and operate without being questioned too closely on how every number was built. That environment is changing, and the shift is happening across multiple directions at once.

Regulatory expectations are becoming more detailed and more consistent. Frameworks like CSRD and BRSR Core are not just asking for disclosures, but for clarity on how those disclosures are constructed. It is no longer enough to report a number. There is an expectation to show the underlying data, the assumptions made, and the methodology used to arrive at it.

Customer and investor demands are becoming more specific. Product-level carbon data, supplier-level transparency, and comparative disclosures are increasingly being requested as part of procurement and investment decisions. These are not one-off requests. They tend to be repeated, refined, and expanded over time, which makes manual workflows difficult to sustain.

Assurance requirements are tightening. As more organisations move toward third-party verification, the focus shifts from producing outputs to defending them. This is where gaps in data quality, methodology, or traceability become much more visible, especially if earlier steps in the process were not fully structured.

Taken together, these shifts are raising the bar in a way that makes the role of AI more relevant, but also more exposed. Faster workflows are useful, but only if they hold up under scrutiny. Systems that produce outputs without making assumptions visible or traceable tend to create more risk, not less.

That is why the conversation around AI needs to become more precise. The question is no longer whether AI can speed up parts of the work. The question is whether the outputs it produces can be understood, explained, and trusted when it matters.

Using AI Responsibly in Sustainability

There is also a layer to this conversation that is easy to overlook. AI is often positioned as a tool to support sustainability work, but it also has its own footprint. The infrastructure behind it - data centres, model training, and continuous processing - requires energy, often at a scale that is not immediately visible to the end user.

That does not make AI contradictory to sustainability efforts, but it does change how it should be used. Applying it indiscriminately, or layering it onto workflows where it does not meaningfully reduce effort, can create additional complexity without clear benefit. In that sense, the same discipline that applies to emissions reduction applies here as well: use it where it creates measurable value, not where it simply sounds advanced.

In practice, responsible use tends to come down to a few simple considerations:

• Use AI where it reduces material effort, not just marginal tasks  

• Avoid duplicating processes that already work without it  

• Be aware of the underlying compute and data requirements, especially at scale  

• Prioritise transparency, so outputs can be understood and reviewed  

These are not strict rules, but they reflect a more deliberate approach. As sustainability teams adopt AI into their workflows, the goal is not just to move faster, but to do so in a way that remains consistent with the broader intent of reducing impact and improving accountability.

How KarbonWise approaches it

The way AI is used in sustainability work tends to reflect how the problem itself is understood. If the goal is speed alone, the focus stays on automation. If the goal is defensibility, the focus shifts to how data is structured, reviewed, and explained.

At KarbonWise, the approach has been to use AI where it reduces friction in the workflow, while keeping human judgement at the centre of anything that affects how results are interpreted or validated. In practice, this shows up not as a single feature, but as a different way of working across the lifecycle of the analysis:

• Faster, more consistent outputs without losing control
  Many of the underlying steps - mapping, structuring, and initial analysis - are supported by AI, which significantly reduces the time required to move from raw data to usable results. The goal is not just speed, but consistency, with assumptions remaining visible and reviewable at each step.  

• Ability to handle highly specific, real-world use cases
  Instead of relying on fixed models, workflows can be adapted to very specific customer requirements. This makes it possible to address smaller, context-specific challenges without rebuilding the system each time, allowing the solution to remain flexible in practice.  

• Shifting effort from building outputs to using them
  The workflow is structured in layers, where an initial output is generated and then reviewed before reaching the user. This reduces the manual effort required at each step, allowing teams to focus less on assembling answers and more on interpreting results, making decisions, and using sustainability data more strategically.  

The distinction is important. In sustainability work, the value of a number is not just in how quickly it is produced, but in whether it can be understood, explained, and trusted when it is reviewed by a customer, an investor, or an assurance provider. AI helps make that process more efficient, but it is the combination of structured workflows and human oversight that determines whether the result is ultimately useful.

AI is clearly becoming part of how sustainability work gets done. The question is not whether to use it, but how to use it in a way that actually improves the quality of the outcome.

In practice, that comes down to being clear about where it adds value and where it does not. It can reduce effort, handle scale, and bring structure to messy workflows. What it cannot do is replace the need to understand the data, question assumptions, or explain results when they are challenged.

The organisations that get the most out of it are not the ones using it everywhere, but the ones using it deliberately. They treat it as a tool that supports the work, not something that defines it.

As expectations around sustainability data continue to rise, that distinction becomes more important. The goal is not just to produce outputs faster, but to produce outputs that hold up - and can be used with confidence when decisions depend on them.

{{cta}}

{{accordion}}

{{sources}}