What a Fully Autonomous NGER Submission Looks Like
Walk through every step of an autonomous NGER reporting workflow, from document ingestion to EERS-ready output. We're honest about which parts work today, which are close, and which still need a human.
There are 978 controlling corporations registered under the NGER scheme right now. Every one of them has the same deadline: 31 October. Every one of them goes through roughly the same process to get there. Collect documents from dozens of sources. Extract consumption figures by hand. Match them to the right NGA emission factors. Calculate per-gas breakdowns. Aggregate by facility. Validate against thresholds. Format for the EERS portal. Submit.
Most of them do this with spreadsheets, shared drives, and a sustainability manager who doesn't sleep much in September.
We've been building Carbonly to automate as much of this pipeline as possible. Not because automation is trendy, but because we spent close to two decades inside enterprise data systems at mining and energy companies and watched this same painful cycle repeat every single year. The science of NGER reporting isn't hard. The logistics of getting clean data into the right shape, on time, with an audit trail? That's where everything falls apart.
So here's what a fully autonomous NGER submission would look like, end to end. And then we'll tell you exactly where we are today, what's close, and what still needs a person in the loop.
The seven-stage pipeline
Picture a reporting year where you don't chase a single invoice. No one opens a PDF to type a number into a cell. No one spends a week reconciling spreadsheets against utility portals. The system just runs.
Stage 1: Data arrives automatically. Your cloud storage sync pulls new files from shared folders weekly. An ingestion email address catches forwarded invoices from accounts payable, site managers, and suppliers. Fuel card providers export transaction data on schedule. Electricity retailers deposit bills into a watched folder. Every document that enters the system gets timestamped and assigned to the right facility and reporting period. Nothing sits in someone's inbox waiting to be processed.
Stage 2: AI reads every document. Each file gets classified (electricity bill, gas invoice, fuel docket, waste manifest, water statement) and processed according to its type. The system reads the document the way a person would, understanding the layout rather than relying on rigid templates. It extracts the quantities that matter: kWh consumed, GJ of natural gas, litres of diesel, tonnes of waste to landfill. Each extraction gets a confidence score. A clear electricity bill with standard formatting might score 98%. A handwritten fuel docket from a remote site might score 72%.
Stage 3: Emission factors get matched. The extracted consumption data gets mapped to the correct NGA emission factor. This isn't a simple lookup. State-based grid factors differ significantly (Victoria's is 0.78 kg CO2-e/kWh versus Tasmania's 0.20). Scope 1 fuels need energy content factors and per-gas emission factors for CO2, CH4, and N2O separately. The system maintains a material library that learns from corrections, so when someone enters "ULP" or "regular unleaded" or "petrol 91" they all resolve to the same factor set.
Stage 4: Quality checks run automatically. This is where most manual processes break down. A quality agent scans every record against a set of rules. Has this invoice been processed before? Does this consumption figure spike more than 40% above the rolling average for this site? Is the billing period overlapping with another record? Are there round numbers that suggest an estimate rather than a meter read? These checks flag exceptions. They don't silently pass bad data through.
Stage 5: Compliance validation. Before anything gets near a report, a validation layer checks that every facility has data for every reporting period. Energy content is calculated correctly. Per-gas breakdowns (CO2, CH4, N2O) are complete for every Scope 1 source. The corporate boundary matches what's registered with the Clean Energy Regulator. If you're above the 25 kt CO2-e facility threshold or the 50 kt corporate group threshold, every required data point needs to be present.
Stage 6: The NGER report gets pre-generated. The system populates the NGER template structure: facility-level emissions by scope and gas type, energy production and consumption, methodology references, and contextual data. It runs its own validation rules against what the EERS portal expects. Any gaps or inconsistencies get flagged before a human ever sees the output.
Stage 7: Approval and submission. A sustainability manager reviews the exceptions, which in a well-running system represent maybe 3-5% of total records. They approve, period locks engage, and the report is ready for EERS submission by an authorised officer.
That's the vision. Seven stages, most of them running without human intervention, producing an auditable NGER report with every figure traceable to a source document.
What actually works today
We don't think it helps anyone to pretend the whole pipeline is fully autonomous right now. Some of it is. Some of it is close. And some of it genuinely needs human judgement for the foreseeable future.
Here's an honest breakdown.
Document processing works well. Carbonly handles eight document formats: PDF, CSV, Excel, Word, PowerPoint, RTF, images, and scanned documents. The AI reads layouts rather than matching against fixed templates, which means it doesn't break when a retailer changes their invoice design. Accuracy improves with use because the material matching system learns from corrections. For standard utility bills, electricity and gas especially, extraction accuracy is high. For messy documents like handwritten fuel dockets or invoices in unusual formats, confidence scores drop and the system routes them for human review. That's the correct behaviour. We'd rather flag something than silently get it wrong.
Material matching covers most scenarios. The five-tier matching system works through NGA database factors, published Environmental Product Declarations, a global emission factor cache, learned aliases, and AI-powered factor lookup. Common materials resolve instantly. Edge cases, things like specific concrete admixtures, niche refrigerant blends, or novel biomaterials, sometimes need a human to confirm the match. We're not sure this will ever be fully autonomous for every material on earth, and we're okay with that. Getting 95% right automatically and flagging the other 5% is dramatically better than getting 72% wrong manually (which is what the ANAO found when they audited NGER report accuracy).
Quality checks are production-ready. Duplicate detection, consumption spike analysis, billing period overlap checks, round number flagging, and missing data alerts all run automatically. We built prebuilt rule sets for construction and corporate reporting because those are the contexts where we see the most volume. These checks catch the kinds of errors that lead to enforceable undertakings. The Clean Energy Regulator accepted one from a major energy producer in July 2025 specifically because their internal controls weren't catching persistent inaccuracies across multiple reporting periods.
Report generation works. The NGER template gets auto-populated from processed data. Pre-validation catches most formatting and completeness errors before you'd encounter them in the EERS portal. We also generate GHG Protocol reports, custom formats, and executive summaries from the same underlying data, which matters because AASB S2 mandatory climate disclosures overlap significantly with NGER data requirements. AASB S2 allows NGER reporters to use their NGER emissions calculations for their climate disclosures, so getting NGER right effectively gives you a head start on ASRS compliance too.
The approval workflow is built. Submit, review, approve, with period locking so approved data can't be accidentally modified. Audit trail captures who changed what, when, and why.
What's close but not finished
Automatic document collection from all sources. Cloud storage sync works. Email ingestion works. But not every supplier portal offers automated exports, and some fuel card providers still require manual CSV downloads. We're building connectors for the most common Australian energy retailers and fuel card providers, but "every source, fully automatic" isn't realistic yet for every business. If your electricity comes from AGL and your gas from Origin, you're probably fine. If you have a diesel supplier in regional WA who faxes invoices, you're going to be forwarding those emails manually for a while.
Auto-confirmation for high-confidence records. Right now, all processed records go into a review queue. We're building a graduated trust model where records above a configurable confidence threshold (say 95%) get auto-confirmed while lower-confidence records require human review. The idea is that a standard AGL electricity bill that the system has processed hundreds of times shouldn't need someone to click "approve." But this needs careful calibration. NGER penalties reach up to $660,000 per contravention at the current Commonwealth penalty unit rate. Auto-confirming records that turn out to be wrong would be worse than no automation at all.
NGA factor auto-updates. The Department of Climate Change, Energy, the Environment and Water publishes updated NGA Factors annually, usually in August, right before NGER reporting season. Ideally the system would ingest these automatically and apply the correct vintage of factors to the correct reporting period. We currently update factors manually when new versions are published. It works. But it's a manual step in what should be an automated pipeline. We'll get there.
What still needs humans
Some parts of NGER reporting require human judgement, and we don't think AI should be making these calls unsupervised. Not yet. Maybe not ever for some of them.
Final sign-off. The NGER Act requires that reports be submitted by an executive officer or someone with EERS "report submitter" permission. This isn't a technology limitation. It's a legal requirement. Someone with authority needs to look at the numbers, confirm they're right, and press submit. That's appropriate. The goal of automation isn't to remove accountability. It's to make sure the person signing off is reviewing a clean, validated dataset rather than a spreadsheet full of question marks.
Scope 3 methodology choices. NGER itself is primarily Scope 1 and 2, but AASB S2 requires Scope 3 disclosure for Group 1 entities (and will extend to Groups 2 and 3). Scope 3 is fundamentally different from Scope 1 and 2 because it involves estimation methodologies, supplier engagement decisions, and boundary judgements that are more policy than calculation. Should you use spend-based or activity-based methods for purchased goods? How do you handle suppliers who won't share emissions data? We can automate the calculations once you've made these decisions, but the decisions themselves need human expertise.
Corporate boundary and structural questions. If you acquired a subsidiary halfway through the reporting year, which months count? If you operate under a joint venture with equity-based allocation, how do you split the emissions? These questions come up more than you'd think, especially in construction and mining where project structures change frequently. The system can apply the rules once someone sets them, but setting them requires understanding the commercial and legal structure of the business.
Unusual materials. If your facility uses a refrigerant that isn't in the NGA Factors workbook, or a proprietary fuel blend, or a novel waste stream, someone needs to research the appropriate emission factor. No database covers everything. We're honest about that.
Why this matters beyond NGER
Here's the thing about building an autonomous NGER pipeline: it prepares you for more than just the 31 October deadline.
AASB S2 mandatory climate disclosures for Group 2 entities kicked in from July 2026. Group 3 starts from July 2027. These disclosures go in your annual report, right next to the financial statements. The data foundation is substantially the same: Scope 1 and 2 emissions, calculated from the same consumption records, using the same (or very similar) emission factors.
If your NGER workflow is automated, your ASRS data is 80% done before you start. If your NGER workflow is a spreadsheet fire drill every October, you're now doing two fire drills. That's not a workload increase most sustainability teams of two or three people can absorb.
The Clean Energy Regulator also isn't standing still. The 2025-26 reporting year amendments introduced market-based reporting for biomethane and hydrogen consumption, updated fugitive emission factors for oil and gas operations, and added new requirements for flared gas reporting. These changes compound. Every year the NGER scheme gets slightly more complex. An automated system absorbs new rules as configuration changes. A spreadsheet process absorbs them as more rows, more formulas, and more opportunities for error.
We're not claiming Carbonly runs a fully autonomous NGER submission today. We're not there yet. But the gap between "mostly automated with human review of exceptions" and "entirely manual" is enormous. It's the difference between your sustainability manager spending September doing data entry and spending September actually reviewing the data.
One of those approaches scales. The other one breaks when AASB S2 Group 3 pulls another few thousand companies into mandatory reporting scope.
If you want to see where your NGER workflow sits on the automation spectrum, reach out at hello@carbonly.ai. We'll walk through your document types, facility structure, and reporting obligations and give you an honest assessment of what can be automated today and what still needs your team. Per-project pricing, no lock-in.