The 2026 Omnibus rollback narrowed mandatory CSRD reporting to the largest companies, those over 1,000 employees and €450 million turnover. It did not remove sustainability data from the economy. It pushed the request one step down the value chain.
The large companies still in scope, and the banks that lend to everyone, still need supplier data, and they still ask for it. A small manufacturer that supplies three large customers can receive three different sustainability questionnaires in a quarter, none of them identical, each with its own spreadsheet and its own deadline. For an SME with no sustainability team, that is a real and recurring cost.
The EU’s answer is the VSME, the Voluntary Sustainability Reporting Standard for non-listed small and medium enterprises. It gives an SME one standard to report against instead of many bespoke forms, and under the new value-chain cap it limits what large customers are allowed to demand. The catch is that most SMEs have no in-house capacity to produce it, and VSME, though far lighter than CSRD, is still a structured disclosure exercise. This article is the AI workflow I would build to get an SME from scattered operational data to a defensible VSME report, and the part where a circular-economy practitioner, not a language model, has to be in the loop.
It is the SME companion to the CSRD workflow architecture. The frameworks differ in scale. The workflow shape is the same.
Key Takeaways
- The requests did not stop, they moved to VSME. The Omnibus cut roughly 85% of companies out of mandatory CSRD, but their large customers still ask suppliers for sustainability data. VSME is where that request now lands.
- VSME is a ceiling, not just a chore. Under the value-chain cap, a company subject to CSRD cannot demand more from an SME supplier than the voluntary standard sets out. Report once against VSME and you can decline the questionnaires that reach above that line.
- Two modules, choose before you start. The Basic Module covers the core most requests need. The Comprehensive Module adds strategy, targets, and detail for banks and investors. The module you pick is driven by who is actually asking.
- AI does the synthesis, not the data or the judgment. Gap analysis against the modules, first-draft disclosures, and mapping VSME to each customer’s questionnaire are where a model helps. Meter readings, HR records, and materiality calls are not.
- The circular-economy disclosures are where a practitioner beats the model. Resource use, waste, and material flows are the part of VSME where domain judgment adds value a model cannot, and the part most worth doing well.
- The adversarial review step still applies. The output goes to your customers and their auditors. Invented figures or a misread threshold are the real risk, not fabricated standards.
VSME in Mid-2026: Who Actually Needs It
The regulatory picture, in plain terms, is this. Omnibus I, adopted in February 2026 and in force from 18 March under Directive (EU) 2026/470, raised the CSRD thresholds to over 1,000 employees and over €450 million turnover. Listed SMEs are fully exempt. Most companies that were bracing for CSRD are now out of mandatory scope entirely.
Being out of mandatory scope is not the same as being free of the question. The companies still in scope have to report on their value chains, and the banks, insurers, and large corporate customers of everyone else continue to send data requests as a condition of finance and supply. That is the pressure VSME is designed to absorb.
The VSME was developed by EFRAG through public consultation and field testing with real SMEs, and the European Commission adopted it as a Recommendation in July 2025. It is voluntary, non-binding guidance, structured in two modules. The Basic Module is a compact set of general, environmental, social, and business-conduct disclosures that the Commission expects to satisfy most requests an SME receives. The Comprehensive Module adds strategy, targets, transition information, and further detail for the deeper requests that come from banks and investors. EFRAG provides the standard and a support ecosystem, including digital XBRL templates.
The compact core: general information, the key environmental disclosures (energy and emissions, pollution, water, resource use and waste, biodiversity), workforce, and business conduct. The Commission expects it to satisfy most requests a supplier receives. Start here.
Everything in Basic, plus strategy, targets, transition information, and additional detail. Aimed at the deeper requests from banks, investors, and large corporates assessing you for finance. Use it only when someone is actually asking for that depth.
The value-chain cap is the part most SMEs miss. A company subject to CSRD is prohibited from requesting information from a smaller company in its value chain beyond what the voluntary standard defines. VSME is that limit. So a VSME report is not only a way to answer efficiently, it is a legitimate basis for declining the parts of a customer questionnaire that reach above the line. That reframes the exercise from an open-ended obligation to a bounded, one-time piece of work.
Where the Bottleneck Actually Is for an SME
The bottleneck is not the standard. VSME is deliberately short. The bottleneck is that a small company does not have the data in the shape the standard wants, and does not have a person whose job is to put it there.
The energy figures are in utility bills. The waste figures are in a contractor’s invoices. Headcount and turnover are in the payroll and accounting systems. Health and safety records are in a folder someone maintains. None of it is wrong, and almost none of it is in a form that maps cleanly onto a disclosure. The work is gathering, reconciling, and framing, then drafting language that a customer’s procurement or ESG team will accept. That synthesis layer is exactly where an AI workflow earns its place, and exactly where it has to be supervised.
The Workflow Architecture
The design is a multi-stage workflow with structured outputs between stages and a human checkpoint at the end, the same architecture as the CSRD version, scaled to the SME’s reality.
Decide Basic or Comprehensive before anything else, driven by who is asking. A supplier answering a mid-market customer needs Basic. A company seeking bank finance usually needs Comprehensive.
Output: the exact target set of disclosures, so nothing is over-reported.
Map the chosen module’s disclosures against what the company already has. The model builds the checklist and flags what is missing. It is forbidden, in the prompt, from inventing a value to fill a gap.
Output: a structured gap report, covered versus missing, per disclosure.
The human, operational stage. Pull the meter readings, the waste transfer notes, the payroll summary. The workflow structures the request so the owner knows exactly which number to find and where.
Output: a tagged evidence pack with a source for every figure.
The model drafts each disclosure narrative from the gathered evidence, in the standard’s structure, with every figure traceable to its source. First draft, not final.
Output: per-disclosure draft narratives with provenance.
Cross-reference the VSME report against each inbound customer questionnaire, so one report answers many forms and the cap is invoked where a form reaches above the line.
Output: a mapping table, one VSME disclosure to many questions.
A second sub-agent argues against the draft. Which figures have no traceable source, which thresholds are misread, which material judgments are presented as settled when they are not.
Output: a verification report flagging items for human resolution.
The key word is structured. Each stage hands the next a structured artefact, a gap report or an evidence pack, not conversational prose the next stage has to re-parse. That is what builds the audit trail in by default and keeps each sub-agent running in a few thousand tokens of focused context rather than one sprawling conversation.
What This Would Look Like as a Claude Code Skill
The workflow above is the design for a Claude Code skill, in the same shape as the consulting skills on the open-source skills hub. A VSME skill would have six components.
Loaded on every session so Claude recognises a VSME task the moment it starts, and knows the inputs it needs: the chosen module and the company evidence pack.
Loaded only when relevant. Holds the six-stage pipeline, the structured-output schema between stages, the value-chain-cap logic, and the adversarial-review prompts.
One per stage, each with its own isolated context, prompt, and output schema. The orchestrator passes structured artefacts between them, not conversational prose.
The VSME Basic and Comprehensive disclosure map, the digital XBRL template structure, and the common shapes of the customer questionnaires the report has to answer.
Every figure in the output links back to its primary source, a bill, an invoice, a payroll line, and every cap decision is logged. The trail is the deliverable, not an afterthought.
The most important and easiest-to-skip part. Prompts that make a second pass hunt for unsourced figures, misread thresholds, and material judgments dressed up as settled.
The architecture follows the same pattern as the consulting skills already on the hub, which means it is buildable today against a real engagement.
Where a Circular-Economy Practitioner Beats the Model
VSME’s environmental disclosures include energy and greenhouse gas emissions, pollution, water, biodiversity, and resource use and waste. The last of those, the material and circular-economy side, is where a model’s limits show and where field judgment matters.
A model can tabulate waste tonnages from invoices. It cannot look at a production process and see where the material actually leaks, which stream is misclassified, where a by-product is being landfilled that has a reuse market, or whether a reported recycling rate reflects real material recovery or energy-from-waste dressed up as diversion. Those are the same judgments I make in circular-economy fieldwork, and they are the difference between a VSME report that satisfies a form and one that surfaces something the business can actually act on. For an SME, that second version is where the exercise pays for itself rather than just costing time.
Where AI Helps and Where It Fails
AI helps most with the parts that are structured and repetitive. Building the gap-analysis checklist, drafting boilerplate disclosures, mapping one report onto many questionnaires, and keeping an audit trail of which figure came from which source. For an SME doing this once a year, that is most of the pain removed.
AI fails, quietly and expensively, on three things. It will fill a data gap with a plausible number if you let it. It will misread a threshold or a definition and apply it confidently. And it will present a material judgment, on what counts as waste, on where a boundary sits, as if it were settled when it is not. None of these are visible in a fluent draft. All of them are the kind of error a customer’s auditor is paid to find.
The Adversarial Review Step Is the Safeguard
The single most important stage is a separate review pass whose only instruction is to attack the draft. Check every figure against its stated source. Flag any number with no traceable origin. Challenge every material judgment. Verify that each disclosure actually maps to the module requirement it claims to.
VSME is voluntary, which can create a false sense that accuracy matters less. The opposite is true. The report goes straight to customers, banks, and their assurance providers, and it becomes the basis on which an SME declines further requests. A number that does not hold up undermines exactly the leverage the report was meant to create. The review step is cheap. The failure it prevents is not.
A Practical Checklist
- Confirm you are actually out of mandatory CSRD scope, then choose VSME Basic or Comprehensive based on who is requesting.
- Run the gap analysis before gathering anything, so data collection is targeted rather than exhaustive.
- Keep every figure traceable to a primary source, a bill, an invoice, a payroll line.
- Have a person, ideally one with domain knowledge, own the resource-use and material-flow disclosures.
- Run the adversarial review before the report leaves the building.
- Use the value-chain cap. Answer VSME once, then point to it when a questionnaire asks for more.
What This Changes for SMEs and Their Advisers
For an SME, VSME turns an open-ended stream of supplier questionnaires into one bounded piece of work with a legal ceiling. An AI-assisted workflow makes that piece of work small enough to do without hiring a sustainability function. For a consultant, the same shift moves the value from filling in forms, which is now largely automatable, to the judgment layer: choosing the module, defending the material calls, and reading the operational reality behind the numbers. That is the part worth paying for, and the part a model cannot do alone.
Frequently Asked Questions
Is VSME mandatory?
No. VSME is a voluntary standard adopted by the European Commission as a Recommendation. What is effectively obligatory for many SMEs is answering their customers and banks, and VSME is the standardised, capped way to do that.
Do I need a consultant, or can our team do it?
Many SMEs can produce the Basic Module in-house with a workflow like this one and a couple of focused days. The case for an adviser is strongest on the material judgments and the resource-use disclosures, and when the report has to withstand a demanding customer or a finance application.
Can I just use ChatGPT for the whole thing?
A single-shot prompt will produce something that looks like a VSME report and quietly invents the numbers it does not have. The point of the multi-stage architecture, with a data-gathering stage and an adversarial review, is to make the failure modes visible before the report leaves your hands. Model choice matters less than workflow shape.
Related reading
The CSRD workflow architecture is the large-company version of this piece, for the firms still in mandatory scope. And if you are weighing which AI tool to build any of this on, the vendor sustainability audits apply the same honesty to the tools themselves.
Turning a supplier questionnaire into a one-time job. I help SMEs and their advisers build VSME and value-chain reporting that holds up, with the material and circular-economy judgment that a model cannot supply on its own. If that is on your desk right now, that is the work I do. See how I work.