Data procurement for corporate carbon footprinting

The mathematical methodology of a carbon footprint is largely standardized. The real bottleneck lies upstream - in extracting reliable consumption data from thousands of invoices, logs and statements - a typical case for AI-supported document processing.

Anyone who has created a carbon footprint knows the picture: hundreds of PDF invoices, electricity bills in a dozen layouts, tank receipts, maintenance logs, freight documents. The actual calculation - quantity times emission factor - is trivial; weeks pass beforehand, collecting, typing and checking plausibility. The majority of the balancing time is not spent on analysis, but on pure data collection.

The EU Omnibus Directive, in force since March 2026, reduces the scope of CSRD reporting obligations - to companies with more than 1,000 employees and a turnover of € 450 million - but leaves the obligation to disclose emissions in place. At the same time, since January 2026, CBAM has required importers of energy-intensive goods to provide verifiable, activity-based data instead of rough estimates.

But it's not just the obligation that motivates people: Anyone who voluntarily writes „climate-neutral by 2035" on their banner also has an obligation to provide proof. The EU Directive 2024/825 (EmpCo), effective from September 2026 via the UWG, and recent BGH case law require clear, verifiable and substantiated environmental claims. Vague, expenditure-based estimates are not enough - what is needed is activity data that backs up every statement right down to the source.

Anyone who communicates ‚climate-neutral by 2035' has a duty of proof. Expenditure-based estimates do not legally secure such a promise - verifiable activity data does.

The eye of the needle: quantities beat sums of money

A carbon footprint according to the Greenhouse Gas Protocol is divided into three areas: Scope 1 (direct emissions, for example from heating and the vehicle fleet), Scope 2 (purchased energy) and Scope 3 (the upstream and downstream value chain). Scope 3 in particular accounts for the lion's share in many industries - up to 90 percent of the footprint.

There are two basic methods of calculation:

• Spend-based: An amount of money is multiplied by an industry-average emission factor. Quick, but inaccurate and price-dependent.
• Activity-based: You use the actual physical quantity - kilowatt hours, liters, kilograms, tonne-kilometres - multiplied by a specific factor. Much more precise and audit-proof.

The GHG Protocol clearly classifies data quality: supplier-specific data first, then activity-based data, then average values and only lastly expenditure-based estimates. The upcoming revision of the Scope 3 standard will even make this difference publicly visible in future by indicating the proportion of reliable versus estimated data.

This is the crux of the matter - and a common misunderstanding. „Physical quantities beat monetary amounts“ initially sounds as if you need data that is not on invoices at all. The opposite is the case: the kilowatt hours are on the electricity bill, the liters on the tank receipt, the megawatt hours on the district heating bill. The reason why so many companies still resort to the imprecise expenditure method is not because the quantities are missing - it is simply too expensive and too slow to extract them manually from heterogeneous documents. It is this bottleneck that AI-supported document processing solves.

Where the data relevant to the balance sheet is contained

It is worth looking beyond the invoice. Quantities relevant to the balance sheet are distributed across a whole range of document types, each with characteristic fields:

• Electricity and gas bills: Consumption in kWh, energy source, billing period (Scope 2 or 1)
• Tank and fuel receipts: Liters or kilograms per fuel type (Scope 1)
• District heating bills: MWh and heat source (Scope 2)
• Refrigerant maintenance logs: refilled quantity in kg and refrigerant type - an often underestimated Scope 1 item due to its high global warming potential
• Bills of lading and CMR: Weight and distance, resulting in tonne-kilometres (Scope 3.4)
• Proof of disposal: Tonnage and type of waste (Scope 3.5)
• Travel receipts, boarding passes, hotel bills: Distances, means of transportation, overnight stays (Scope 3.6)

One special case deserves precision: the Energy certificate. It is not a consumption record in the strict sense, but provides the energy demand or consumption per square meter, the primary energy and the energy source of a building. For the operational balance of a building in use, the actual consumption is taken from bills and meters. The energy performance certificate comes into its own when it comes to real estate portfolios, the EU taxonomy or approximate values in the absence of consumption data.

What the data must do

Before a figure can be included in an auditable balance sheet, it must be more than just a value. It must unitized (250 what? kWh or MWh determines the factor), correctly assigned in terms of time (accounting periods rarely run exactly with the financial year) and referable - i.e. traceable back to the source document. The latter is not a convenience, but a prerequisite for the verification security required under CSRD and for the upcoming data type labeling of the GHG Protocol. An extraction that only provides a bare numerical value is worthless for balancing; what is required is the value plus unit plus energy source plus time period plus supplier plus source reference.

AI software as a reliable first mile

Intelligent Document Processing (IDP) fulfills precisely this requirement profile. AI software such as Konfuzio processes an incoming document in four steps:

• Classify: The system automatically recognizes whether a document is an electricity bill, a consignment note or an energy certificate - even with mixed batches.
• Extract: The fields relevant to the balance sheet are read from the document, including quantities from continuous text and tables.
• Validate: Units are normalized, values are checked for plausibility. If the detection reliability is below a defined threshold, the system escalates the case to a human (human-in-the-loop) instead of silently guessing.
• Handed over: The structured data is sent to the target system, the accounting software, via REST API or webhook

The clear delineation of responsibilities is crucial. IDP provides the structured, unit-based activity variable including source reference. The Assignment of the emission factor and the actual greenhouse gas calculation is performed by the carbon footprint software. „Solving with Konfuzio“ therefore does not mean calculating the balance sheet, but rather maximizing the proportion of activity-based usable data - and thus moving the company up a level in the data hierarchy.

It is not the simple, one-page invoice that makes the difference, but the difficult case: the multi-page document with nested tables, several consumption points and individual items that extend over page breaks. Here comes classic OCR reaches its limits because it reads layout but no meaning. However, it is precisely these documents that often contain the most quantities - and therefore the biggest source of errors if a position is missing or two values are mixed up. Reliable extraction must understand such structures, check their completeness itself and escalate in case of doubt instead of skipping lines unnoticed. The care invested here pays off later in the check.

Three data pools for the carbon footprint

So that the expectation is correct, a sober classification helps. IDP can only extract what is documented. The data of a carbon footprint is divided into three pots:

• Document-based - the core area of IDP. Large parts of Scope 1 and 2 as well as several Scope 3 categories are located here. It is the most expensive block manually and therefore offers the greatest leverage.
• System and API data - Meter readings, ERP master data, telematics, travel management tools, supplier EDI and the increasing exchange of supplier data via standards such as Catena-X or PACT. This often does not flow through the document route at all and is better connected via integration. However, if a supplier sends its product carbon footprint as a PDF, this is another case for IDP.
• Modeled or raised - Commuting by employees, the use phase of products sold, some process emissions. This is not documented and cannot be extracted.

IDP makes the document-based activity data exhaustively accessible - far beyond the invoice - but does not replace the system and model data, which come from other sources anyway.

The handover - and the resulting use cases

The practical benefits are decided at the transfer point. Using a REST API, the extracted, verified data records are transferred directly to a carbon footprint or carbon management software. The market for this is mature and differentiated; Tanso, Plan A, Cozero, ClimatePartner and Greenly, among others, are established in the DACH region and internationally. They all rely on an input variable that they can rarely obtain cleanly themselves: structured, verifiable activity data. The interface therefore closes a real gap - not between any two tools, but between the unstructured document world and the structured accounting logic.

This coupling results in use cases that would be almost impossible to manage manually:

• Continuous instead of annual balance sheet: Receipts are processed continuously, the footprint is updated on a rolling basis instead of at the end of the year.
• Deepen Scope 3: Bills of lading and proof of disposal open up categories that are otherwise estimated on a flat-rate basis.
• CBAM imports: document-based quantities for the mandatory reporting of energy-intensive imports.
• real estate portfolios: Structured evaluation of energy performance certificates in existing buildings - for balance sheet, assessment and taxonomy at the same time.
• Supplier PCF onboarding: automatically transfer incoming footprints as a PDF to the balance sheet software.
• Greenwashing prevention: Voluntary climate targets and PR statements can be backed up with evidence-based activity data - every value can be traced back to the source document in the event of queries from competitors or environmental associations. 

The benefits can be soberly quantified, even without marketing promises: If the effort shifts from manual entry to exception processing, the role of the sustainability team also shifts - away from typing and towards analysis and a reduction strategy. In addition, there is a less visible but key quality for the audit: each value remains linked to its source document. If the auditor asks, a figure can be traced back to the original invoice in seconds instead of searching for it in folders. The same seamless traceability that the auditor requires for the CSRD assurance also works externally: if a voluntary climate pledge is requested from competitors or environmental associations, every figure can be traced back to the original document. 

Conclusion

In practice, climate accounting rarely fails due to methodology or calculation logic. It fails because of the data collection - the laborious, error-prone manual work between incoming mail and accounting software. This is exactly where intelligent document processing comes in: it retrieves the physical quantities from the documents in which they are already stored, checks and structures them and transfers them in a traceable manner to the systems that turn them into a balance sheet. This does not replace balance sheet expertise or the other data sources. But it frees climate accounting from its most expensive bottleneck - and turns estimated values into reliable figures.

Do you need reliable data for your climate assessment? Please contact Contact to us. Our experts will give you a non-binding initial assessment.