AI in CSRD: what the people doing the actual work are saying

Every leadership team has the same vision about AI and CSRD: reporting gets faster, cheaper, and maybe, if we’re lucky, almost fully automated over time. Earlier this year, we tested that hypothesis against reality. 

Nordic Sustainability facilitated a roundtable with sustainability and reporting leaders from several leading companies in our network. These experts represent leading European companies in manufacturing and retail, and know the ins and outs of sustainability reporting.  

We put one question to them: Where in the CSRD process is AI already a helpful tool, and in what part of the process are the human experts still invaluable?   

Here’s what the companies we spoke to are actually doing with AI, and where the efficiency dream runs into the reality of the work.

The DMA: AI as research partner, not judge 

The Double Materiality Assessment (DMA) is where the clearest early wins sit. Desktop research, benchmarking competitor reports, and analysing stakeholder interview transcripts: AI handles these well, and many teams are already using it to speed them up.  

Some are exploring next-generation DMA tools built directly on the latest versions of the European Sustainability Reporting Standards (ESRS). 

The bigger ambition is a DMA that’s a living document: continuously updated from new science, regulation, and business events, flagging when a new market or product launch should shift the company’s Impacts, Risks and Opportunities (IROs) picture, without waiting for the annual cycle. That’s already within reach.   

AI speeds up the research behind the DMA, but scoring IROs is still a human call

What AI doesn’t do is score IROs against the business as well as the people who know the business do. Understanding the company, agreeing how material risks and opportunities should be framed, and connecting them credibly to the business model still takes expert judgment.  

AI can inform the scoring, but the decision has to be owned by someone who knows what they’re looking at. 

Data: Data quality is the real roadblock 

Even the most advanced companies in the room still struggle with the quality of their sustainability data. AI is now exposing errors in datasets that previously went unnoticed. Garbage in, garbage out – just at scale and with more visibility. 

AI is exposing data problems companies didn’t know they had 

For AI to be genuinely useful in this process, data quality and infrastructure need to improve first. The companies getting the most out of AI in this phase are the ones that already did the unglamorous work of cleaning and structuring their data.  

Until that problem is fixed, AI’s powers can’t fully be harnessed. 

Report writing: Draft fast, then stay critical 

CSRD reporting is a big exercise, and a pattern emerged across the group: an expert drafts quickly, then uses AI to make the output more precise, consistent, and audit-friendly. This applies to reporting manuals, audit-friendly documentation and the final output. First-pass checks against disclosure requirements already save time. 

But the efficiency only holds if the reviewer knows enough to catch what AI gets wrong. That critical eye can’t be switched off.  

The output also usually needs humanising, since it’s too polished in the wrong way, or not how the company actually writes. That’s a prompting discipline and not a fundamental limit, but it takes intentional effort to fix. 

Should companies disclose what AI wrote? 

One question is being asked but not yet answered: should companies disclose which text was written by AI versus a human? The more mature organisations in the group were already thinking about it. 

Consensus was that we will likely see partially AI-generated CSRD reports very soon, but how many will be transparent about it? 

Assurance: AI as sparring partner, not a stand-in (at least not yet) 

Limited assurance under CSRD demands time, effort and patience, especially the first time round.  

Some of the companies we spoke to are already using AI in audit preparation: stress-testing arguments before they go to auditors and drafting audit memos quickly when the underlying documentation is solid.  

Company context and interpretation still need a human 

Who knows, maybe in a few years it will be companies’ internal AI running the assurance conversation with the assurer’s AI agents. For now, humans are still needed for the harder discussions about company context and interpretation that are also part of the audit cycle.   

What the freed-up time is actually for 

CSRD reporting has brutal peaks: months of evenings and weekends, then quieter stretches. AI that meaningfully cuts that burden creates space. The more interesting question is what fills it. 

Most companies still treat sustainability reporting as a compliance exercise rather than a strategic tool. That shows up clearly in how many material topics are only partially backed by concrete targets or incentive schemes. A long list of material topics inflates disclosure volume and data collection effort, drives up the cost and complexity of reporting, and pulls focus from the issues that actually matter. 

The time AI frees up should go towards decision-making and impact, not just cost-cutting 

The time AI frees up is a chance to repair the link between what companies report as material and what they actually do about it: deeper internal stakeholder engagement, working through what material IROs actually mean for strategy rather than just documenting them, and improving the quality of Scope 3 data rather than just collecting more of it. 

This is what most sustainability departments dream of doing, and this is also where the true business value lies – reducing risks and spotting opportunities that make the business stronger. 

Whether that happens, or whether the saved time simply equals a lower headcount, is an open question. In some organisations, near term, it may be the latter. But sustainability work doesn’t end. Regulation keeps coming, risks keep growing, and the companies that redirect freed-up time towards decision-making and impact will be better placed than those that just cut costs. 

AI’s own footprint is a dilemma too  

There’s a cost to all this efficiency that also needs to be made transparent. Training and running large models draws heavily on water and energy for data centre cooling, and the hardware behind them carries its own human rights concerns, from mineral sourcing to labour conditions in AI supply chains.  

For people who’ve built a career on reducing environmental and social impact, pushing AI adoption without naming that trade-off feels uncomfortable, even hypocritical. It’s a genuine dilemma that many of the experts in the field are struggling with.  

Should AI’s footprint be treated as a material topic in CSRD reporting? 

This inevitably leads to the question of whether AI’s own footprint should be treated as a material topic in CSRD reporting. The experts we had gathered leaned towards yes: it makes sense, given how fast AI is being adopted across corporates with almost no visibility into its environmental and social cost. So far, very few companies actually report on it. 

What we took away 

AI is already being used across CSRD reporting today. Not evenly, and not yet with a clear roadmap. Maturity varies sharply even among the most advanced companies, and so does the transparency on how AI is leveraged. 

Leadership’s dream is speed and saved time/money, while the people doing the actual work tell a more nuanced story: expert judgment is still needed at almost every stage, from IRO scoring to audit interpretation, and it isn’t going away. And poor data is still the biggest efficiency roadblock. 

AI can make reporting work faster and more streamlined, but doesn’t replace human expertise. For sustainability departments (most of which still remain chronically understaffed and wearing hundreds of hats), the dream of AI in sustainability reporting is about freeing time for the work that drives genuine sustainability and business impact, which also is what ultimately brought most of us to this field of work in the first place.  

That freedom comes with its own responsibility: using AI deliberately, not just because it’s there, and being honest about what it costs as well as what it enables.  

This article was written with AI support and carefully reviewed by a human expert – just like your sustainability report very soon will be.