A sustainability manager at a mid-size French industrial I spoke with last month keeps a private log of how she spends her days. Over a normal quarter she logged 71% of her hours on what she calls "moving numbers" — opening a supplier PDF, finding a figure, typing it into Excel, copying the result into a regulator's portal, then emailing the supplier to explain that the figure was in the wrong unit. She has two master's degrees. She is, by training, a materials scientist. By function, she is a data-entry clerk with a LinkedIn headline that says "ESG & Climate Lead."
That is the thesis of this piece. Sustainability compliance, as currently practiced, is roughly 20% expertise and 80% clerical work — and the clerical 80% has been misclassified as expertise for long enough that we've built an entire consulting industry around billing for it. The future of AI sustainability compliance is not a prettier dashboard or a richer template library. It is an AI agent that reads your documents and fills the forms, so the humans can go back to the 20% that actually required a degree.
The expensive fiction of "ESG expertise"
Walk into any compliance team at a mid-cap and the workflow is the same. An email arrives from a supplier in Hebei at 2 a.m. local time with a scanned PDF, in Chinese, of a factory's energy consumption for Q3. Somebody prints it, somebody else retypes the kilowatt-hours into a column in a workbook named CBAM_tracker_v17_FINAL_use-this-one.xlsx, a junior analyst converts kWh to MWh and then doubts themselves and converts it back, and by Friday the number is sitting in three different places with two different values and a half-finished comment bubble that reads "double-check with Wei."
None of that is ESG expertise. Rewriting a number into a second spreadsheet is not sustainability work any more than photocopying a contract is legal work. It's the plumbing underneath the work. And compliance teams have been drowning in it because the regulation explicitly requires that the plumbing be traceable, auditable, and filed in a format the regulator invented in 2023 and has revised twice since.
The numbers on this are brutal enough to quote. Internal surveys at two CBAM declarants we've worked with put the share of compliance time spent on "data movement" at 63% and 78%. A CSRD working group at a listed Dutch manufacturer clocked it at 71%. The 2025 IFAC practitioner survey on sustainability assurance readiness puts the figure across its sample at 68%. Call it two-thirds, conservatively. Two-thirds of the budget you think you're spending on sustainability strategy is being spent on retyping.
And here's the part that should embarrass the industry: the data entry is now being performed by some of the most overqualified people in the building. Ex-McKinsey associates. PhDs in environmental engineering. A friend of mine with a doctorate in life-cycle assessment spent three weeks last November reconciling unit mismatches between a German supplier's technical data sheet and a Czech refinery's material balance. That is data entry wearing a lab coat.
Why the clerical work grew, not shrank
The obvious response is: surely software has been solving this for years. Carbon platforms exist. ESG platforms exist. Sphera, Workiva, Persefoni, Watershed, Novisto — none of them are nothing. A lot of money has gone into this category.
It has, and the clerical burden has gone up anyway, for a reason that is structural rather than about any one vendor's product decisions. The platforms are built around forms. The forms have fields. The fields expect clean, typed, validated, pre-mapped input. None of your source data arrives in that shape. Your source data arrives as a scan, a photo of a meter, a faxed bill of lading, a CSV with Cyrillic headers, an email body where your supplier's sales rep has pasted numbers into the body because he doesn't know how to attach a file.
The platform sits on one side of a gap. The source documents sit on the other. A human being — usually an expensive one — stands between them and manually moves the data across, field by field. The platform has improved the landing strip. It has not touched the runway. That is why the time spent per filing has not meaningfully dropped in a decade, even as the number of vendors has quadrupled. You cannot fix a document-to-form translation problem with a better form.
What would actually fix it is a system that reads the document and writes the form. Which, until roughly 2024, was science fiction at scale.
Where this already works
CBAM is the honest test case, because the forms are brutal and the source documents are chaos. A quarterly CBAM filing requires mapping each imported good to a CN code, identifying direct and indirect emissions by production route, tracking precursor materials, logging carbon prices paid at origin, and producing an XML file that the EU's Transitional Registry will accept without complaint. The source material is supplier production data sheets, test reports, utility invoices, internal mass balance calculations — in whatever language and format the supplier happens to produce. A consultant will charge you €10,000 to €50,000 per quarter to translate one into the other. A platform will give you a faster spreadsheet.
An AI agent does something categorically different. You hand it the supplier's PDF, in Chinese, and it extracts the coal-vs-gas fuel mix, matches the product description to a CN code, cross-references the EU emission factor database, flags that the electricity consumption looks low for the stated output volume, asks you a clarifying question, and produces a filled CBAM quarterly template with every field linked back to the exact page and line in the source document. First filing, in one evening. Second filing, in ninety minutes, because it remembers last quarter's product mix.
The GHG Protocol inventory is the same shape of problem and the same shape of solution. Scope 1 is fuel invoices, Scope 2 is utility bills, Scope 3 is the procurement ledger and a swamp of supplier attestations. Every single one of those inputs is a document. The "work" of building a GHG inventory is 15% methodology judgment (how do I categorize this leased asset, which emission factor set applies in Mexico) and 85% extracting numbers from documents and putting them in the right cells. Of those two halves, the second is the one that software can now do and couldn't before.
This is the shift. It is not the one the vendor landscape has been pricing. Most ESG platforms are still organized around "help your humans fill the form faster." The shift is: the humans stop filling the form.
Steelmanning the other side
The serious objection to all of this goes roughly like this. Compliance isn't data entry. Compliance is judgment. Deciding whether a supplier's carbon price counts for CBAM's carbon-price-paid deduction is a legal question, not an OCR question. Deciding which topics are material under ESRS double materiality is a board-level conversation, not a document-extraction task. Deciding whether to disclose a contested Scope 3 methodology that might embarrass a key customer is a reputational call that requires a human being with skin in the game. Hand any of those to an AI and you will, at best, get a confidently wrong answer presented with the same calm tone as a correct one. At worst you'll get an audit finding that costs more than the consultant fee you were trying to avoid.
All of that is right. I'd go further: it is more right than the AI-evangelist version of this argument usually concedes. Double materiality under ESRS is not a data-extraction problem. It is a stakeholder-engagement problem layered on a risk-assessment problem layered on a legal-disclosure problem. The AI can draft the IRO matrix. It cannot decide which impacts your board is willing to put on the cover of the annual report. Contested materiality, legal disclosure carve-outs, novel methodological interpretations, defending a number under cross-examination from a Big Four auditor — these are jobs for humans, and they should stay jobs for humans, and they should be paid as jobs for humans, which they often currently aren't because the fee budget has been eaten by the clerical 80%.
The point is not that AI replaces sustainability expertise. The point is that AI replaces the data-plumbing pretending to be sustainability expertise, and in doing so it finally lets the real expertise surface. When a materials scientist isn't spending 71% of her week retyping supplier PDFs, she can spend that week doing the one thing that actually requires a materials scientist, which is looking at the production data and asking whether the factory's reported emissions intensity is plausible given the technology it's running. That question pays for her salary. Retyping doesn't.
What the near future looks like
Concretely, the compliance function three years from now looks like this. One person, with domain expertise, oversees a stack of frameworks — CBAM, CSRD, GHG Protocol, CDP, EU Taxonomy, ISSB, a few sector-specific ones like SBTi commitments or LCA screenings — that used to take a team of five. The person does not open spreadsheets. The person uploads documents into an agent and reviews what the agent produced. The agent handles extraction, mapping, unit conversion, cross-framework reuse (the same supplier energy number feeds four different reports; it gets typed once), validation, and the production of whatever file format the regulator asked for this week.
The person still does materiality. The person still makes the legal calls. The person still sits in the assurance meeting and defends the methodology choices. But the person does not do data entry anymore, because the data entry is not a job a human should ever have been doing.
This is what Formist is — an AI compliance platform built by WeCarbon that reads your source documents and fills the forms, across CBAM, GHG Protocol, CSRD/ESRS, EU Taxonomy, CDP, ISSB, SBTi, LCA, and a dozen more. It isn't a better spreadsheet. It's the end of the spreadsheet being the unit of work. It won't decide your materiality for you. It will make sure the 400 data points underneath your materiality decision arrive at the form pre-filled, with every number linked to the source document it came from.
The quotable line
Compliance should be about judgment, not transcription. The consulting industry got rich pretending those were the same job. They're not, and an AI agent that reads documents and fills forms has just made the distinction impossible to hide.
Stop filling forms. Go do the work you were hired for.
Formist is built by WeCarbon, a climate-tech company with offices in Paris, Shanghai, and Dubai. It supports CBAM, GHG Protocol, CSRD/ESRS, EU Taxonomy, CDP, ISSB, SBTi, LCA, and 15+ other sustainability frameworks.