Precision Carbon Intelligence for Global Enterprises

Enterprise carbon accounting is a data engineering problem wearing a sustainability label. Emissions are activity data multiplied by emission factors, under the methodology of the GHG Protocol. The arithmetic is trivial; assembling trustworthy activity data, versioning the factors, and producing a calculation trail an assurance provider can follow is not.

#From Voluntary Reporting to Audited Disclosure

What this means for you: the methodology has not changed in twenty years. What changed is that the output now sits inside audited corporate reporting, which makes it an engineering problem with your name on it.

For two decades, corporate carbon reporting has been anchored by the GHG Protocol Corporate Standard — the de facto global methodology defining the three emission scopes, boundary rules, and calculation approaches that nearly every framework and regulator references.

In the EU, the Corporate Sustainability Reporting Directive (CSRD) and its European Sustainability Reporting Standards (ESRS) require in-scope companies to report climate information, including greenhouse gas emissions, within formal corporate reporting and subject to assurance. In the United States, California's SB 253 requires large companies doing business in California to disclose Scope 1 and Scope 2 emissions, with Scope 3 following on a later phase.

The engineering consequence of the word "audited" is the important part.

An assurance provider does not accept a number. They accept a derivation.

Every reported tonne needs a traceable path back to source records, the factor applied, the method chosen, and the person or system that made each judgment call. That requirement, not dashboard aesthetics, is what should drive the architecture — and it is the same data lineage discipline that governs financial reporting systems, applied to a domain that has historically been run out of spreadsheets.

#The Three Scopes, Treated as a Data Problem

What this means for you: each scope has a different data supply chain, a different quality ceiling, and therefore a different investment case. Treating them as one programme is how budget goes to the wrong place.

Scope 1 — direct emissions from sources the organisation owns or controls: on-site fuel combustion, company vehicles, fugitive emissions from refrigerants and industrial processes. The activity data lives in fuel invoices, fleet telematics, and maintenance logs.

Scope 2 — purchased energy: electricity, steam, heating, and cooling generated elsewhere and consumed by the organisation. The GHG Protocol requires dual reporting under two methods: location-based (the average emission intensity of the grid where consumption occurs) and market-based (reflecting contractual instruments such as power purchase agreements and energy attribute certificates). The two numbers can differ materially, and both must be supportable.

Scope 3 — value chain emissions, organised by the GHG Protocol into fifteen categories spanning upstream (purchased goods and services, transportation, business travel, employee commuting) and downstream (use of sold products, end-of-life treatment, investments). In most non-energy-intensive enterprises, Scope 3 represents the large majority of the footprint — and it is calculated substantially from other organisations' data, which defines its quality ceiling.

The quality column is the strategy: a credible programme moves up the data-quality ladder from spend-based estimates towards activity-based and supplier-specific data, wherever footprint concentration justifies the effort.

#The Data Engineering Problem Underneath

What this means for you: three problems consume most of the engineering effort, and none of them are the calculation. Budget accordingly.

#Activity Data Fragmentation

The inputs live everywhere: ERP purchase orders, utility portals, travel booking systems, fleet tools, facilities spreadsheets, and — for Scope 3 — supplier questionnaires and third-party datasets.

They arrive in inconsistent units (kWh, therms, litres, tonne-kilometres, currency), granularities (meter readings versus annual invoices), and latencies. The unglamorous work of entity resolution, unit normalisation, and gap identification is the same discipline as any enterprise data integration programme, which is why our data engineering teams treat a carbon platform as an integration project with a reporting deliverable rather than the reverse.

#Emission Factor Management

Emission factors — the coefficients that convert activity into CO2-equivalent — are reference data with all of reference data's failure modes. They come from national and institutional libraries (government conversion-factor datasets, grid-intensity databases, life-cycle inventories), vary by country and region, and are revised on publication cycles.

A factor library therefore needs versioning, effective-date handling, and source citation on every factor. When a factor or methodology changes, prior-year figures may need recalculation under base-year policy — tractable only if the calculation layer can replay history against a new factor version.

Treating the library as governed reference data with a named owner is the difference between an estate that can replay a prior year on request and one that rebuilds it from memory every reporting cycle.

#Scope 3 Supplier Data Quality

Scope 3 begins as spend-based estimation: multiply procurement spend by an industry-average factor per currency unit. It is a legitimate starting method and a poor ending one.

Maturing means collecting primary data from high-contribution suppliers, validating their submissions as rigorously as internal data, and tracking the footprint share covered by each method tier so the improvement is itself reportable.

#Reference Architecture: Ingestion to Audit Trail

What this means for you: four stages, each with a contract to the next. If you can name the owner of each contract, the platform is assurable; if you cannot, it is a spreadsheet with better styling.

Two of those stages carry most of the risk. In factor mapping, unmapped or ambiguously mapped records must go to a review queue rather than to a default factor.

Silent defaulting is where footprints quietly go wrong.

In the audit trail, the design goal is narrow and testable: an assurance engagement should be a review rather than an excavation. Analytics, target tracking, and forecasting sit on top of this pipeline as consumers of it, not substitutes for it — and the continuous quality checks that keep the inputs honest are the same ones described in continuous data observability, operationalised through our DQ Sentinel service.

#Where Machine Learning Helps, and Where It Does Not

What this means for you: machine learning belongs inside this architecture as an accelerator, never as the thing that produces a reported figure.

What machine learning does not do is repair missing primary data or make an estimate audit-grade. A model-imputed value is still an estimate and must be labelled as one. Forecasts inherit the uncertainty of their inputs, and assurance providers evaluate traceable calculations rather than model outputs.

The honest pattern is machine learning as an accelerator inside a deterministic, auditable pipeline — flagging, classifying, projecting — with reported figures produced by transparent calculation. Where we build these models through our ML and predictive analytics practice, the classification output is always a suggestion routed to a review queue, never a value written straight into a reported total.

#Honest Limitations

What this means for you: state these caveats in the report yourself. An assurance provider or an activist analyst will find them anyway, and finding them first is the cheaper position.

• Scope 3 double counting is inherent. One company's Scope 1 is another's Scope 3; value-chain totals across companies are not additive, by design. Scope 3 figures are decision-useful for prioritisation rather than precise allocations.
• Factor uncertainty bounds everything. Industry-average factors carry wide uncertainty ranges, and a footprint built on them inherits those ranges regardless of the decimal places shown.
• Market-based instruments are contested. Whether certificates and offsets represent real, additional reductions is an active methodological debate; dual Scope 2 reporting exists so readers can see the footprint both ways.
• Comparability between companies is limited. Boundary choices, method tiers, and factor sources differ legitimately between reporters, so cross-company comparisons often measure methodology choices as much as performance.
• Data maturity takes years. Moving Scope 3 from spend-based to supplier-specific data is a multi-year supplier-engagement programme sequenced by footprint concentration, not a platform feature.

Precision in presentation is not accuracy in measurement.

Reporting the method mix honestly while it improves is a stronger position than reporting a single number to two decimal places. It is also the position that survives a second year, when the figure moves because the method improved rather than because the emissions did.

That distinction is one your data strategy and analytics functions should be able to evidence on request, using the same policy-as-artefact discipline described in governance as code.

#Frequently Asked Questions

#What are Scope 1, 2, and 3 emissions?

They are the GHG Protocol's three categories of corporate emissions. Scope 1 covers direct emissions from sources the company owns or controls, Scope 2 covers emissions from purchased energy, and Scope 3 covers all other value-chain emissions upstream and downstream, organised into fifteen categories. For most non-energy-intensive enterprises, Scope 3 is the large majority of the footprint and the hardest to measure.

#How do location-based and market-based Scope 2 differ?

The location-based method applies the average emission intensity of the grid where electricity is consumed. The market-based method reflects the company's contractual arrangements, such as power purchase agreements and energy attribute certificates. The GHG Protocol's Scope 2 guidance calls for reporting both, and they can diverge substantially — which is precisely why both are required rather than either alone.

#Do we have to report Scope 3 emissions?

Increasingly yes, for companies in scope of the major regimes. CSRD and ESRS reporting covers material value-chain emissions, and California's SB 253 phases in Scope 3 disclosure after Scopes 1 and 2. Thresholds and timing differ by regime and have shifted during implementation, so confirm current regulator guidance for your entity. The data supply chain takes the longest to build, so plan ahead of the date.

#How accurate does carbon accounting need to be?

Accurate enough to be assurable and decision-useful, not laboratory-precise. Assurance providers evaluate whether the methodology is sound, consistently applied, and traceable rather than whether every tonne is exact. The practical standard is proportionality: metered precision where meters exist, transparent estimation with stated methods where they do not, and an improvement path that puts better data on the largest categories first.

#What does an assurance provider actually check?

Whether the reported figure can be derived again from evidence. In practice that means the boundary and consolidation approach, the completeness of reporting entities, the provenance and version of each emission factor, the method tagged on each result, the treatment of estimates and manual adjustments, and the approval trail behind them. A platform that can replay a prior period on request answers most of this without a special exercise.

---

Unolabs is a Data and AI first engineering consultancy, headquartered in the United Kingdom with engineering operations in Pune and active engagements across the UK, Australia, and Hong Kong. We help enterprises build the architectural foundation for autonomous AI execution — governed data platforms, semantic intelligence, and agentic systems that enterprises can stand behind.

If you are building a carbon platform that has to survive an assurance review rather than a board slide, book a discovery call.