AI Invoice Carbon Engine for Australia

Many carbon accounting tools import a dollar figure from your accounting system and multiply it by an industry-average factor. Under AASB S2 assurance, that shortcut is hard to defend in a walk-through when the activity data was in the invoice all along, and it cannot show a reduction when you buy lower-carbon materials at the same price. Carbonly is an AI invoice carbon engine built in Australia: it is designed to read the tonnes of concrete on a batching-plant delivery docket, the tonne-km on a freight consignment, the kWh on an electricity invoice and the litres on a fuel docket, match each to the correct NGA 2025 emission factor with confidence scoring, and link every kilogram of CO2-e back to the exact page of the source document.

Talk to us

What changes for ai invoice carbon engine

Physical activity data, not dollar spend, as the default input for Scope 1, 2 and 3 - the AASB S2 assurance walk-through has something real to trace
Eight file formats in one engine: PDF, CSV, Excel, Word, PowerPoint, RTF, images, and scanned or image-only PDFs from phone photos in a truck cab
Alias learning: when a reviewer corrects a match, the line text is saved as an alias for your organisation, so the next docket with the same wording matches through it
Five-tier material matching to NGA factors or a construction EPD factor library, with confidence scoring on every match and low-confidence matches sent to a reviewer
Credit lines with negative quantities reduce the total, and fee, hire, time and dollar-only adjustment lines are skipped rather than turned into emissions

What the system does

AI reads supplier invoices from Boral, Holcim, InfraBuild, Cleanaway, Veolia, SUEZ, and other construction and waste suppliers
AI reads freight invoices from Toll, Linfox, StarTrack, and other logistics providers with tonne-km on the consignment
AI reads electricity bills from AGL, Origin, EnergyAustralia, Alinta, Red Energy, and other retailers across the states
AI reads fuel invoices from Ampol, BP, Shell/Viva Energy, Caltex, and the regional fuel suppliers your subcontractors use
AI extracts emissions automatically from unstructured PDF documents, structured CSV exports, spreadsheet workbooks, and phone photos
Deterministic anomaly checks flag outlier fuel consumption and unit mismatches daily at 6am and after each document batch, and the Data Health Agent flags coverage gaps, incomplete records and stale data
Provenance chain persisted from the disclosure line, through the calculation, through the emission factor version applied, through the material match rationale, through the reviewer who confirmed it, back to the original supplier PDF with a content fingerprint

Frequently asked

How accurate is the AI invoice extraction across different suppliers?

Publishing a single global accuracy number is misleading because accuracy varies by supplier, document type, and document quality. What Carbonly does instead is expose the confidence score on every match and send low-confidence matches to a reviewer. When a reviewer corrects a match, the line text is saved as an alias for your organisation, so the same wording matches next time. Accepting a match that was already right does not add an alias.

What about handwritten fuel dockets and phone-photo receipts?

Eight file formats include image-only PDFs and photos taken on a phone in the field. Where OCR confidence is low, the row lands in review with a "low confidence - please confirm" tag. If the reviewer corrects the material match, the line text is saved as an alias for your organisation, so the same wording matches through it next time.

Does the AI ever post an emission record without human review?

High-confidence matches are recorded with their confidence score. Low-confidence matches, and units that cannot be converted without a figure the document does not state (for example cubic metres to tonnes with no density), go to a reviewer rather than straight to the ledger. Every ledger write is recorded as a typed audit event.

What if we want to keep spend-based Scope 3 for some categories?

Every emission record carries a methodology label - activity-based, hybrid, supplier-specific, average-data, or spend-based. Categories where activity data genuinely does not exist (Category 15 investments for a non-financial firm, Category 11 use-of-sold-products for many industries) legitimately fall back to spend-based or average-data, with the methodology label preserved on the record. Under AASB S2 paragraph 29(a)(vi) disaggregation by category, the reporter can see and defend the methodology mix.

Ready to see how this works for your operation?

Email hello@carbonly.ai with your reporting obligation, number of sites or projects, and current process. We reply within business hours.