The AI Document Extraction Workflow in Carbonly: A Screenshot-Led Walkthrough
Supplier invoices and dockets turned into ledger records you can trace back to the page they came from. A screen-by-screen walk through upload, review, ledger, dashboard and the NGER report, using screenshots of the product.
What a reporting team gets from Carbonly is simple to state. Supplier invoices, fuel dockets and utility bills become emission records, and each record links back to the document it came from. This post shows that working, screen by screen.
Every screenshot below is from the product, captured from a test workspace with sample data. The organisation and facility names in them are test data, not a client.
If you are evaluating carbon accounting software, the useful question is this. What does the screen look like when a reviewer has to sign off on an emission record that may later be assured?
Every number traced back to a source document
Under AASB S2 and an ASSA 5010 assurance engagement, we expect an assurance provider to ask how each disclosed figure was produced. That usually means a source document, a stated method, the emission factor used and a review record. NGER record-keeping points the same way. A number nobody can trace to a document is hard to defend.
So the screens worth looking at are the plain ones. The review queue. The match against a factor. The link from a ledger record back to its invoice.
Step 1: Get documents in without changing how suppliers send them

The upload screen takes one file or a batch. You pick the project the documents belong to, then upload and extract. Carbonly reads eight file formats: PDF, Word, PowerPoint, Excel, CSV, RTF, image files and scanned images.
Most operational data does not arrive one file at a time, so upload is one of three ways in. The second is OneDrive or SharePoint sync, which pulls new files from a folder you point it at. The third is email ingestion. Projects can carry their own ingestion address, visible in the projects screen further down, so invoices sent or copied there land in the right project.
Step 2: Review the extracted line items before anything reaches the ledger

This is the screen that matters most. The source file sits on the left. The extracted line items sit on the right, each matched to a material and an emission factor.
Each match shows a percentage, and the queue can be filtered by confidence band. Low-confidence matches go to a reviewer. The matching runs through a five-tier cascade against NGA factors or a construction EPD factor library. We wrote up how in five-tier material matching for carbon accounting.
Two warnings sit above the list. One flags lines whose units cannot be converted for the matched material, so they will not post until the unit or material is fixed. The other flags lines that may duplicate existing emission records and deselects them by default.

The reviewer can change the material on any row before confirming. When a reviewer corrects a match, or assigns a material to a line with no match, the line text is saved as an alias on that material. Later documents with the same wording match through it. Aliases stay inside your organisation and are capped at 500 per material.
Accepting a match that was already right does not add an alias. The review step stays in place for every batch. We describe the human-in-the-loop design in AI carbon accounting with self-learning and human-in-the-loop accuracy.
Extraction is not perfect on a first pass. Handwritten dockets and low-resolution scans are the usual hard cases. That is why the review queue exists and why low-confidence lines go to a person.
Step 3: Confirmed lines become records an auditor can follow

Once a reviewer confirms a line, it becomes an emission record in the ledger. Three things on that record matter for assurance.
First, the basis of the figure. A record built from a physical quantity (litres, kWh, GJ, kg) is treated differently from one built from spend. Spend-based Scope 3 records are floored to low data quality, because spend estimates carry wider uncertainty. Fee, hire, time and dollar-only adjustment lines are skipped. Credit lines with negative quantities reduce the total. More on this in AI document processing for fuel dockets and utility bills.
Second, the factor version. Each record carries the emission factor version used to calculate it. When factors are updated, you can see which version sits behind each historical figure.
Third, the source document. Each record links to the document it came from, and the verbatim line text is kept as evidence. An auditor's sample request becomes a lookup, not a search through an archive.
When a reporting period is closed, period locking stops further edits. A change after that needs a restatement reason, which is recorded.
Projects that match how your reporting boundary is drawn

Projects are how Carbonly groups records. A project can be a site, a facility, a business unit or a joint venture stake. The choice depends on how your boundary is drawn.
Projects can sit in a hierarchy, as the screenshot shows, and carry an ownership percentage. Joint venture consolidation supports operational control, financial control and equity share.
Access runs on six organisation roles: owner, admin, manager, contributor, auditor and viewer. Suppliers get a separate role through the Supplier Portal, where they submit data scoped to a project. Multi-factor authentication is available.
Most NGER reporters are not single-facility. A construction contractor may run many sites, each with different concrete and fuel suppliers. Rolling those up for an NGER submission works best when the site-level records stay intact underneath.
A dashboard built on the same records

The dashboard reads from the same confirmed records as the reports. The screenshot is a test workspace, so its percentages and trend lines reflect sample data, not a real organisation.
Two panels on it are worth knowing about. The anomaly count comes from deterministic rule-based checks. They run daily at 6am and after each document batch. They flag records that fall outside expected ranges, so someone can look before the period closes.
The Data Health panel checks completeness: missing scopes or months, incomplete records and stale data. It is a separate check from the anomaly rules. Custom dashboards can also be set up and sent by scheduled email.
An NGER report you can check line by line before entering it in EERS
The NGER report is the output most Australian reporters ask about first.

One point needs to be clear. The CER's EERS portal does not take file uploads, and the screen says so. Carbonly's NGER report is a record-keeping document and a reference for entering figures into EERS. It is not the lodgement itself.

The report summary shows total Scope 1, total Scope 2, energy in GJ and the number of facilities. Below that is a per-gas breakdown using AR5 GWP values, which is what NGER uses. Threshold status is computed from the ledger against the corporate group thresholds of 50 kt CO2-e or 200 TJ of energy. The facility thresholds are 25 kt CO2-e or 100 TJ.
The report also has tabs for facilities, activity data, a validation check, exclusions and approval. It exports as XML, PDF, CSV or JSON. The XML follows the EERS structure. It is a reference file, and the reporting entity stays responsible for what it lodges.

Facilities are registered with their EERS ID, state, ANZSIC code, control status and grid. Records roll up to these facilities for the report.
The Australian National Audit Office's performance audit of the NGER scheme reported that, of 545 NGER reports it reviewed, 72 per cent contained errors and 17 per cent contained significant errors. The report's validation tab is a check to run before entering figures. It does not replace a person checking the method and the boundary.
For the 2025-26 year, the NGER deadline is midnight AEDT on Monday 2 November 2026, because 31 October falls on a Saturday. The CER has said there are no extensions. Our NGER deadline preparation guide covers the last-month checks.
What this means for the person who signs off
Extraction feeds the work. Review is where the figure becomes defensible. The confidence bands, unit and duplicate warnings, material corrections and alias learning are the working surfaces.
The ledger is the single source for the dashboard, the NGER report, the AASB S2 disclosure export, the Auditor Workspace evidence pack and the anomaly checks. That is what lets an assurance provider follow a disclosed number back to a source document.
Consultants who prepare client submissions can use the same workflow. Client staff can be seated as contributors, and an external reviewer as an auditor with read-only access. The consultant keeps the judgement calls on method and boundary.
If you are evaluating platforms in earnest, a practical test is your own documents. Send a few fuel dockets, an electricity invoice and a concrete delivery docket to hello@carbonly.ai. We will run them through the review workflow with you on a call. Pricing is a per-project licence with a A$100 a month workspace minimum and no per-seat fees.
FAQ
Can I try the platform before signing up? Yes. There is a free trial, and we can run your own documents through a workspace with you. Pick invoices where you already know the right answer, so you can check the extraction and matching yourself.
What file formats does Carbonly read? Eight formats: PDF, Word, PowerPoint, Excel, CSV, RTF, image files and scanned images. Documents can be uploaded, synced from OneDrive or SharePoint, or sent to an email ingestion address.
How does a reviewer confirm or correct an extraction? Each line shows its matched material and a match percentage. The reviewer can change the material on a row, then confirm the selected lines. A correction saves the line text as an alias for your organisation, so the same wording matches later.
Does Carbonly lodge the NGER report for me? No. EERS does not accept file uploads. Carbonly produces the report as a record-keeping document and a reference for entering figures into EERS. Your organisation lodges it and stays responsible for it.
What if the extraction is wrong? Low-confidence matches go to a reviewer, and lines with units that cannot be converted are held until fixed. The reviewer corrects the material before confirming. Ledger writes are recorded as typed audit events, and each record keeps its source document and line text.
Related reading
- AI document processing for fuel dockets and utility bills
- AI carbon accounting with self-learning and human-in-the-loop accuracy
- Five-tier material matching for carbon accounting
- NGER report generation automation
- NGER October 2026 deadline preparation guide
- AASB S2 climate disclosure report generation
Related articles
- Carbonly Co-Pilot: Every Question You Can Ask Your Carbon Ledger in Plain English
- The Emission Data Module: How Every Kilogram of CO2e Actually Gets Onto the Ledger
- Carbon Planning and MACC Curves: How Australian CFOs Prioritise Emission-Reduction Capex
- Configurable AI for the Carbon Lead: Org Instructions, Supplier Templates, Calculation Rules, and Custom Formulas Explained
- The Auditor Workspace: Giving Your ASSA 5010 Assurance Team a Self-Service Walk-Through