Your Sustainability Team Is Doing Data-Clerk Work....

Picture a typical Australian sustainability team in 2026. Two to five people. A head of sustainability, maybe a climate analyst, perhaps an ESG coordinator, and if the business is lucky, a data person who was borrowed from finance. Between them, they cover anywhere from 30 to 400 sites.

Now picture what they actually did last Tuesday.

One of them opened an electricity bill PDF. Typed the kWh into a spreadsheet. Opened the next one. Typed the kWh. Opened a fuel docket. Typed the litres. Opened a concrete delivery ticket. Tried to work out whether "32 MPa" belonged in the material column or the notes column. Moved on. 180 more to go before lunch.

That is data-clerk work. It does not reduce a single tonne of emissions. It does not prepare the board for AASB S2. It does not build a Safeguard Mechanism strategy. And it is where most of the week goes.

The honest picture of a sustainability team's week

We talk to sustainability leads across construction, mining, property, and mid-market manufacturing. The pattern is almost identical.

Week one of the month: chase site managers and suppliers for documents. Week two: extract numbers from those documents into a spreadsheet or upload them into a SaaS UI that somehow still requires you to type the number yourself. Week three: chase the gaps, reconcile the obvious errors, argue with accounts payable about whether the August gas bill is in the August or September period. Week four: attempt analysis. Fail. Write a status update for the CFO. Start again.

There is an industry estimate that 60 to 80 percent of corporate emissions reporting effort is data collection and processing. We think that understates the problem for Australian mid-market reporters, because it measures the work that gets done rather than the work that doesn't. Supplier engagement, scenario analysis, board briefings, variance explanations, Safeguard baseline strategy. None of it happens when your two best people are typing litres into rows.

Why compliance is moving against manual extraction

Here is the uncomfortable part. The regulatory direction in Australia is not "sustainability reporting has become easier." The direction is toward more granular, per-line-item, audit-ready evidence. Manual extraction does not scale into that world. It breaks.

Three specific pressures are changing the shape of the work.

AASB S2 and ASAE 3410 assurance. Group 2 reporters go live for financial years starting 1 July 2026. Group 3 follows a year later. The auditor does not want a spreadsheet with a total. They want a traceable chain from the source document to the emission record, to the factor applied, to the AR6 GWP used for the disclosure, to the number in the financial report. Remember that NGER still uses AR5 GWP values while AASB S2 requires AR6, so the same facility can produce two legitimate numbers for two regulators in the same reporting year. Your spreadsheet needs to handle that cleanly, and for most teams it doesn't.

Activity-based Scope 3. Retailers, banks, and ASX200 buyers are pushing past spend-based estimates. They want quantities. Litres of diesel, tonnes of concrete, kilograms of refrigerant, vehicle-kilometres, tonne-kilometres. Spend-based Scope 3 survives where the data isn't there, but procurement-driven disclosure requests are increasingly asking for the physical activity number. That means extracting quantities from supplier invoices, not just running a spend report out of the ERP.

NGER enforcement. The ANAO audit of NGER reporting found 72 percent of sampled reports contained errors, with 17 percent classified as significant. Criminal penalties for dishonest reporting run up to two years imprisonment. Records must be kept five years from end of reporting year. When the Clean Energy Regulator asks for evidence behind a 2024-25 figure in 2028, "I remember typing that number in from a fuel docket somewhere" is not the answer you want to give.

Compliance does not care whether your team is working hard. It cares whether each line item is evidenced, traceable, and defensible. A team of three typing figures into spreadsheets cannot produce that for 200 sites.

What agentic actually means for carbon accounting

The word "agentic" gets thrown around loosely. In a carbon accounting context we use it specifically, and we think it's worth being clear about.

An agent in this setting is not a chatbot that answers questions. It is not a workflow that runs when you press a button. It is software that watches for work, does the work, and raises a flag when something is off, without a human starting it each time.

We have five of them running in production.

The Data Health Agent runs a weekly seven-check health scan across your emissions data. It looks for things that would embarrass you in an audit. Missing periods, duplicates, factor mismatches, material assignments that look wrong against historical patterns. It surfaces them before the auditor finds them.

The Trust Graduation Agent watches how often a particular material type is correctly auto-matched. When the system has seen a supplier's diesel invoices enough times and the confidence has held, it graduates that material to auto-confirm. Your team stops reviewing the routine and starts reviewing the unusual.

The Cross-Document Correlation Agent groups related evidence. A fuel docket, a purchase order, and an invoice that all refer to the same delivery get linked together, so when the auditor asks for the chain behind a line item, the chain is already assembled.

The Narrative Intelligence Agent generates monthly, quarterly, and AASB S2 prose reports with an audit trail. The numbers are attached to the sentences. If a disclosure says "Scope 1 emissions increased 4.2 percent quarter on quarter," the sentence is traceable to the records that made it true.

The Report Readiness Agent tracks NGER and AASB S2 deadline readiness proactively. It knows what is missing, how many business days you have, and what will trip the deadline if it stays missing.

These agents read source documents and create emission records. Your team stops typing and starts reviewing. That is the reframing that matters.

From filling forms to scanning activity

The product industry has spent a decade building carbon SaaS that is essentially a data-entry form with dashboards bolted on. A user uploads or emails a document. A human reads the document. A human types the numbers into a form. A chart appears.

We think that is the wrong shape. If you are still typing numbers into a UI, the software has not actually automated your work. It has just repainted the spreadsheet.

The Carbonly flow is different. Documents arrive through per-project email ingestion, native OneDrive sync, native SharePoint folder sync, or direct upload in PDF, CSV, Excel, Word, or images. The agents read them. Materials are matched against the NGA library, your custom factor library, or the global cache. Emission records are created. The audit trail is immutable, with every record linked back to the source document it came from.

Activity-based quantities are the default where they exist. We prefer litres, tonnes, tonne-kilometres, kWh, vehicle-kilometres, FTE. Spend-based is the fallback, not the starting point. This matters because most tools default to spend-based for the wrong reason, which is that they cannot extract quantities from invoices. When you can extract quantities, you should.

One working example worth mentioning. We are currently engaged with a Tier 1 Australian construction company, starting at a single site, to prove out the agentic approach before portfolio rollout. The deliberate sequencing is the interesting part, not any claimed result. You prove the chain end to end on one site first, then scale.

What "focus on what matters" actually looks like

If your team stops being clerks, what do they do instead? This is the question nobody asks, and the answer is what justifies the entire reframing.

Supplier engagement. Real conversations with your top 20 emitting suppliers about decarbonisation roadmaps, primary data sharing, and whether the EPD they sent through is product-specific or an industry average. This is where Scope 3 reduction actually happens.

Scenario analysis for AASB S2. Paragraphs 29(f) and 33-36 require you to tell a coherent story about climate resilience under different warming scenarios. That is analytical work. You cannot do it while typing litres.

Safeguard Mechanism baseline strategy. If your facility is in the net, you need a multi-year plan for the 4.9 percent annual baseline decline, including where ACCUs, SMCs, electrification, and fuel switching fit. This is a board-level strategy conversation, not a reporting task.

Board briefings and variance explanations. When emissions go up 6 percent quarter on quarter, the board asks why. "I'll pull together some numbers next week" is not an answer. "Here is the three-site driver analysis with the two specific material contributions" is an answer.

Data quality governance. Deciding which suppliers graduate to auto-confirm, which anomalies are real versus noise, how to handle the AR5 to AR6 transition in parallel disclosures. Judgement work, not typing.

This is the work a sustainability team was actually hired to do.

What to do about it

We will not pretend there is a clean switch. Moving from clerical extraction to agent-assisted reporting takes a transition, and the people doing the current clerical work often know the data better than anyone. They should be running the review, not replaced by it.

Two practical starting points. First, measure honestly how much of your team's week goes to data extraction versus analysis. If extraction is above 40 percent, you have a structural problem that a better spreadsheet will not fix. Second, pick one high-volume, low-judgement document type such as electricity bills or diesel dockets, and get that flow to zero-click end to end before you try to do everything.

If you want to see how the agents actually run against your documents, email hello@carbonly.ai and we will set up a working session with your real records. Not a sanitised demo.

Your Sustainability Team Is Doing Data-Clerk Work. Compliance Doesn't Care.