Carbon Inventory for Companies That Operate in More Than One Place
Emission factors were never the hard part. Collecting defensible data from twelve sites, four countries and three ERP systems is. Lessons from building GreenScope, a multi-site carbon inventory platform for CSRD/ESRS reporting.
Most carbon accounting tools assume a single entity with tidy data: one legal structure, one ERP, one energy contract, one person who owns the numbers. Real enterprises look nothing like that. They have twelve sites, four countries, three ERP systems, and a facilities manager at each location who keeps the electricity bills in a drawer.
The hard part of carbon accounting was never the emission factors. Those are published, versioned, and boring. The hard part is collecting defensible data from people whose actual job is something else — and producing a number an auditor will sign off on.
Why multi-site carbon inventory breaks standard tools
A single-entity tool models the world as one spreadsheet. A multi-site group has to answer harder questions before a single kWh is converted into CO2e:
- Which legal entity owns which site, and does the group report on an operational-control or equity-share basis?
- Which country's grid emission factor applies, for which reporting year, and which factor version was in force when the invoice was issued?
- What happens when a site changes energy supplier mid-year, or a lease starts in March?
- How do you reconcile an ERP export in one country with a scanned PDF invoice in another?
- Who is accountable when a number changes after the report was approved?
None of those are calculation problems. They are data model and workflow problems — which is exactly why they get solved with spreadsheets and email until an auditor asks for evidence.
What CSRD/ESRS actually asks for
Under CSRD and the ESRS E1 climate standard, a group has to report Scope 1, Scope 2 (both location-based and market-based) and material Scope 3 categories, consolidated across the whole reporting boundary, with the methodology disclosed and the figures subject to limited assurance.
"Limited assurance" is the part that changes everything. An assurance provider does not just want a number — they want to trace it back to a source document, see which emission factor was applied, and confirm that nobody silently edited a value after sign-off. That means an audit trail is not a nice-to-have feature; it is the product.
The data model that makes it work
The pattern that survives contact with real enterprises separates four things that spreadsheets mash together:
- Organisational structure: group, legal entities, sites, and the consolidation basis — versioned over time, because acquisitions and disposals happen mid-year.
- Activity data: the raw quantity (kWh, litres, tonne-kilometres) with its source document, period, and the person who submitted it.
- Factor set: which emission factor library, which version, which region and which reporting year — pinned to the calculation, never to the current default.
- Calculated results: derived values that can always be recomputed from the three layers above, never edited directly.
If a result can be edited by hand, the inventory is not auditable. If a factor is applied at read time from a global default, last year's report changes when the library updates. Both are common and both fail assurance.
The real bottleneck: collection, not calculation
A facilities manager in Cluj has a full-time job that is not sustainability reporting. If your data collection asks them to log into an unfamiliar platform, understand Scope 2 market-based accounting, and pick the right unit from a dropdown of forty, you will get a blank field and a follow-up email in April.
What works instead:
- Ask for the artefact they already have — the invoice, the meter reading, the fuel card export — not a derived number.
- Accept the messy channel: email attachment, WhatsApp photo, shared drive folder. Extraction is your problem, not theirs.
- Pre-fill everything you already know: site, period, supplier, expected range. A reviewer confirming a value is ten times faster than a contributor entering one.
- Flag anomalies against the site's own history, so a decimal-point error surfaces in week one, not during the audit.
- Show each site its own completion status. Nothing drives submission like a visible gap next to a colleague's green tick.
This is where document extraction and AI agents earn their place in a carbon platform — reading invoices, normalising units, matching a supplier name to a known entity, and routing exceptions to a human. Not to guess emissions, but to remove the typing.
GreenScope: from re:solved project to company
GreenScope started as a re:solved project and became a company. It is a multi-site carbon inventory platform built for CSRD/ESRS reporting, and I own the technical side: the architecture, the data model, the build, and the infrastructure it runs on.
The same constraints we apply to every client system apply there: data stays where the client's compliance team needs it to stay, every number traces back to a source document, and the automation reduces human work without ever becoming an unaccountable black box.
Related reading: Is it safe to give your customer data to an AI? · What to look for in an AI automation agency · AI automation services — managed roles
Frequently asked questions
Why can't we just use spreadsheets for multi-site carbon accounting?
You can, until you need assurance. Spreadsheets don't record who submitted a value, which emission factor version was applied, or what changed after sign-off. An assurance provider asks for exactly those three things, and reconstructing them a year later is far more expensive than capturing them as you go.
What makes multi-country reporting harder than single-country?
Different grid emission factors per country and per year, different invoice formats and languages, different ERP systems, and different fiscal or lease boundaries. The calculation is identical; the reconciliation is where the work is.
Where does AI help in carbon accounting?
In collection and normalisation: extracting quantities from invoices, matching suppliers to entities, converting units, and flagging anomalies against a site's history. It should never invent an emission factor or estimate a missing value silently — every derived figure has to trace back to a source.
Does CSRD apply to us?
Scope depends on size thresholds and listing status, and the phase-in has shifted. If you're a large EU group, or a non-EU group with significant EU turnover, assume you're in scope and check the current thresholds with your auditor — the data collection work takes longer than the reporting itself.
Related reading
- What to Look For in an AI Automation Agency (+ Red Flags)
A practical buyer's guide for founders: what to ask, what to verify, and how to tell a partner that runs your operation from a vendor that ships a slide deck.
- AI Automation for Healthcare: HIPAA, GDPR, and the Self-Hosted Option
Clinics want less admin, not more risk. Here's how self-hosted AI agents automate scheduling, intake, and follow-ups while keeping patient data HIPAA- and GDPR-compliant.
- Make.com vs Zapier for AI Agents: Which Platform Builds Autonomous Workflows?
Make.com and Zapier both connect apps. But for AI agents — systems that decide, remember, and act on their own — architectural differences matter more than brand.