AI for ESG Reporting
Scope narrowed sharply in 2026, the waves were removed, and the organisations still in scope mostly paused when they should not have.
Sustainability reporting was substantially rewritten in 2026, and the most common mistake we now see is a company still in scope behaving as though it is not.
AI for ESG reporting covers data lineage and consolidation across source systems, framework and standard mapping, evidence retrieval for assurance, disclosure drafting from verified data, gap analysis against applicable standards, and peer and benchmark analysis.
What the 2026 changes actually did
Omnibus I was adopted on 26 February 2026 and entered into force on 19 March 2026. It is a substantial narrowing, and being precise about it matters more than reacting to the headline.
- CSRD now applies above €450 million net turnover and 1,000 employees. The phased wave system was eliminated entirely, and transposition is due by 19 March 2027.
- CS3D moved to €1.5 billion turnover and more than 5,000 employees, with transposition by 26 July 2028 and application postponed to 26 July 2029.
- Revised ESRS are to follow by delegated act — fewer data points, priority on quantitative reporting, clearer mandatory and voluntary distinction, and sector-specific binding standards dropped for non-binding guidance.
- Assurance was softened. Limited assurance standard adoption deferred to 1 July 2027, and the reasonable assurance requirement removed.
- CS3D due diligence was restructured into scoping on reasonably available information followed by targeted in-depth assessment, with mandatory climate transition plans and the harmonised civil liability regime removed.
Being out of scope is a determination, not an assumption
The most damaging position we encounter is a company that paused its reporting programme when the simplification was announced and has not since established where it sits against the new thresholds. If you are above €450 million net turnover and 1,000 employees, the requirement did not go away and the transposition date is 19 March 2027 — meaning the data work has less runway than the original waves allowed, not more. A half-day scoping determination resolves this and is worth doing before any other decision.
Data lineage, which is what assurance examines
The reported figure is the easy part
Consolidating a disclosure is arithmetic. Demonstrating where each number came from, how it was transformed, what estimation was applied and who approved it, is the work — and it is what an assurer tests.
Sources are more numerous than anyone expects
Facility meters, HR systems, procurement, logistics providers, supplier submissions, spreadsheets maintained by individuals, and prior-year figures carried forward, all feeding one figure by different routes.
Estimation must be visible in the lineage
Where a number was estimated, extrapolated or gap-filled, the method and the assumption belong in the record rather than in someone's memory of last year's process.
Restatements need to be traceable
Methodology and factor updates change prior figures, and a lineage that cannot reproduce last year's number on last year's basis makes every restatement an argument.
Framework mapping, drafting and the retrieval problem
| Application | Fit | Note |
|---|---|---|
| Data lineage and consolidation | Strong | The core requirement and what assurance examines |
| Framework and standard mapping | Strong | Same underlying data point serving several frameworks |
| Evidence retrieval for assurance | Strong | Finding the document behind a number, in minutes |
| Gap analysis against standards | Strong | What is required, what exists, and what is missing |
| Disclosure drafting | Good | From verified data with every claim traceable; reviewed before publication |
| Peer and benchmark analysis | Good | Useful context; be careful about comparability across methods |
| Policy and commitment consistency checking | Strong | Claims against evidence, across documents |
| Automated disclosure without review | No | Greenwashing exposure and assurance both rule it out |
Evidence retrieval is the capability assurance teams notice
During assurance, the recurring cost is finding the document, calculation or approval behind a specific figure — a request that routinely takes days and generates internal escalation. A retrieval system over the reporting evidence base, with provenance on every result, turns that into minutes. It reduces the assurance burden directly, it makes the internal review cycle faster, and it is a capability that pays for itself in the first reporting cycle rather than in a projected benefit.
How an engagement runs
Determine scope first — it is a half-day and it decides everything after it.
Scope determination and gap analysis
Where you sit against the current thresholds, and what is actually required of you.
Lineage mapping
Every reported figure traced to source, with estimation and approval recorded.
Build
Consolidation with lineage carried through, or evidence retrieval across the reporting base.
Trial
Through a reporting cycle and an assurance walkthrough, measured on request turnaround.
Operation
Lineage maintained as sources change, restatements traceable to basis.
What you receive
A reporting process where every number can be traced, quickly.
Scope determination
Where the organisation sits against current thresholds, documented.
Data lineage
Every reported figure traced to source, transformation, estimation method and approval.
Consolidation
Across facility, HR, procurement, logistics and supplier sources into one reported set.
Framework mapping
One data point serving multiple frameworks without separate collection.
Evidence retrieval
The document, calculation or approval behind any figure, found in minutes.
Restatement traceability
Prior figures reproducible on their original basis when methods change.
Is this the right starting point?
Worth being direct. There are situations in ESG reporting where custom AI work is the wrong spend, and those are listed rather than buried.
Worth doing if
- You paused reporting preparation and have not determined scope against the new thresholds.
- Assurance requests for evidence behind a figure take days.
- The same data point is collected separately for several frameworks.
- Estimation methods live in people's memory rather than in the record.
- Restatements are contested because prior figures cannot be reproduced.
Do something else if
- You want disclosures drafted and published without review.
- Source systems cannot be accessed and lineage is the objective.
- The goal is a more favourable number rather than a more traceable one.
- Nobody will own the lineage after the reporting cycle ends.
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
Are we still in scope after the Omnibus?
That is a determination worth making explicitly rather than assuming, and it takes about half a day. CSRD now applies above €450 million net turnover and 1,000 employees, with the phased wave system removed and transposition due by 19 March 2027. CS3D applies above €1.5 billion turnover and more than 5,000 employees, with application postponed to 26 July 2029. The position we see most frequently is a company still comfortably in scope that paused when the simplification was announced — which leaves less runway than the original waves allowed, not more.
What should we build if we are in scope?
Data lineage, because it is what assurance actually examines. Consolidating a figure is arithmetic; demonstrating where each number came from, how it was transformed, what estimation was applied and who approved it is the real work. Sources are more numerous than anyone expects — facility meters, HR systems, procurement, logistics, supplier submissions, individual spreadsheets and prior-year figures carried forward — and the lineage is what turns a reported number into a defensible one.
How can AI help with assurance?
Evidence retrieval, mostly, and it is the capability assurance teams notice immediately. The recurring cost during assurance is finding the document, calculation or approval behind a specific figure, which routinely takes days and generates internal escalation. A retrieval system over the evidence base, with provenance on every result, turns that into minutes. It reduces the assurance burden directly and pays back within the first reporting cycle rather than as a projected benefit.
Can AI draft our disclosures?
From verified data, with every claim traceable and a review before publication — and not unattended. Assembling descriptive sections from your own verified figures is a genuine time saving in a process with severe deadline pressure. What rules out automation is the combination of greenwashing exposure and assurance: a claim that overstates, or that the evidence base does not support, is a regulatory and reputational problem regardless of how it was produced. The drafting is automatable; the accountability is not.
Do the revised standards mean less work?
Less reporting work, not necessarily less data work, and the difference matters. The revised ESRS are to remove less-important data points, prioritise quantitative reporting and drop sector-specific binding standards for non-binding guidance, which reduces the disclosure surface. The underlying requirement to know where your numbers come from, and to have them assured from 1 July 2027 under the deferred limited assurance standard, is unchanged. Organisations that read the simplification as a reason to stop building data infrastructure have misread which half was simplified.
Related verticals
Organisations in ESG reporting usually share data, buyers or regulators with these. All fourteen are listed on the Agriculture, Food & Environment page.
AI for Climate and Sustainability
Emissions data engineering, supplier extraction and physical risk — with estimated data labelled as estimated.
Read more →AI for Carbon Markets
Remote sensing MRV, project screening and permanence analysis — with methodology chosen for certification, not for volume.
Read more →AI for Food and Beverage Brands
Demand and promotion forecasting, origin data and retail execution — with the deforestation obligation treated as a sourcing question.
Read more →Tell us what the problem looks like.
Thirty minutes, no charge, no deck. We will tell you whether this is an AI problem, a data problem, or a process problem — and we will say when the honest answer is to buy something rather than build it.