How Crab.AI uses artificial intelligence to automate carbon inventories

The problem: collecting ESG data by hand

In most companies, collecting data for an emissions inventory is a manual, fragmented, slow process. The sustainability analyst has to request information from dozens of areas, facilities, logistics, procurement, HR, finance, and each replies in a different format: Excel sheets, PDF invoices, ERP reports, emails with attachments, legacy systems.

The typical result: weeks of work to consolidate data that then has to be reviewed line by line, because typos, swapped units and duplicates are unavoidable. Every inventory cycle repeats the same effort. And when the auditor asks to trace where a figure came from, the analyst has to dig through email folders and spreadsheet versions.

That is the problem Crab.AI solves. Instead of humans extracting data from documents, the AI does the reading, extraction, classification and validation, and the human focuses on analysis and decisions.

How Crab.AI reads documents (energy bills, freight documents, HR reports)

Crab.AI is the artificial intelligence module of the Mangue Tech platform. It processes documents in multiple formats, PDF, image, Excel, CSV, invoice XML, and extracts the data relevant to the emissions inventory.

Electricity bills. Crab.AI automatically identifies consumption in kWh, contracted demand, contract type (regulated or free market), the distributor and the reference period. For companies on the free market, it also identifies renewable energy certificates (I-RECs) tied to the contract.

Fuel invoices. It extracts volume (litres), fuel type (diesel S10, gasoline C, ethanol, CNG), the supplier's tax ID and the date. For fleets, it consolidates consumption automatically by vehicle or cost centre.

Travel reports. It processes statements from travel agencies and booking platforms. It identifies origin, destination, cabin class, airline and date for each leg. It calculates flight emissions using ICAO and Defra factors.

Logistics data. It reads Brazilian electronic freight documents (CT-e, MDF-e) and TMS reports. It extracts cargo weight, origin, destination, mode and distance travelled. For carriers with their own fleet, it cross-references refuelling data.

HR data. It processes commuting surveys and consolidates them by transport mode, average distance and frequency. It integrates home-working data to adjust commuting emissions.

Waste data. It reads waste transport manifests, weighing reports and final-disposal certificates. It classifies by waste type and destination (landfill, recycling, incineration, composting).

The process is transparent: each extracted data point is linked to its source document, with a timestamp, a confidence level, and the option of human review for low-confidence cases.

Automatic classification of spend by GHG Protocol category

One of Crab.AI's most powerful capabilities is classifying spend data automatically by GHG Protocol category.

Many companies have consumption data scattered across ERP and finance systems. The information exists, but it is not classified through the lens of an emissions inventory. A payment to a carrier shows up as "freight" in the ERP, but in the inventory it could be Category 4 (upstream transport) or Category 9 (downstream transport), depending on the flow.

Crab.AI analyses the expense description, the supplier's tax ID, the accounting code and the cost centre to classify it automatically into one of the 15 Scope 3 categories, or into Scopes 1 and 2. The classification uses models trained on real inventory data and is refined continuously.

For companies starting from a spend-based approach, this classification is particularly valuable: it produces a preliminary Scope 3 inventory from financial data already on hand, with no extra collection.

Emission factor suggestions and anomaly detection

Once activity data is extracted and classified, Crab.AI automatically suggests the most appropriate emission factor for each record.

The suggestion takes into account: activity type (combustion, electricity, transport, waste), geography (Brazil, state, municipality), reference period (grid factors vary by year), source priority (MCTI over IPCC, IPCC over Defra) and GWP version (AR5 or AR6, as configured for the inventory).

The platform keeps more than 60,000 emission factors up to date. When MCTI publishes an update, the factors are incorporated and inventories can be recalculated automatically.

Anomaly detection. Crab.AI compares each data point against historical patterns and sector benchmarks. If a site's diesel consumption doubles from one month to the next with no apparent explanation, the system flags it for review. If an energy bill reports consumption inconsistent with the size of the site, an alert fires.

That automatic validation is essential to quality. Manual inventories often carry errors that go unnoticed until audit: swapped units (litres vs gallons), duplicates, overlapping periods, stale factors. Crab.AI catches those patterns before they contaminate the result.

Audit trail: how every figure stays traceable

Auditability is a design principle of Crab.AI, not a feature bolted on later.

Every figure in the final inventory has a complete provenance chain: source document (PDF, invoice, report), extracted data (consumption, volume, distance), GHG Protocol classification (scope, category), emission factor applied (source, version, GWP), the calculation performed (formula, result in tCO2e), who validated it and the date of the last review.

That chain is preserved in full in the system. When the auditor questions a specific figure, say freight emissions at the Manaus site in March, the analyst can navigate from the total emissions to the source invoice in three clicks.

The audit trail also records changes: if a data point was corrected, who corrected it, when and why. That meets the verification requirements of the GHG Protocol, CDP, IFRS S2 and CVM 193.

Comparison: manual process vs Crab.AI

The most significant difference is not speed, it is reliability. An inventory produced with Crab.AI has native traceability, real-time validation and methodological consistency between cycles. That sharply reduces the risk of rework at audit and raises stakeholder confidence in the results.

Crab.AI is available as an integrated module of the Mangue Tech platform. Activation is progressive: a company can start with one document type (energy bills, say) and expand as it gains confidence in the automated process.