Carbon Measurement data methodology

For every transaction sent via the API, we follow a spend-based approach for estimating CO2e (Carbon Dioxide Equivalent). The amount of CO2e, in kg, released from every transaction is:

kg CO2e = Emission factor × Spend amount

For every transaction we select the appropriate emission factor for the spend category, then adjust the spend for taxes, margins, and inflation.

The scope of emissions included is cradle-to-gate plus use — all emissions embodied in the purchased product from its creation, plus the direct emissions from its use. In GHG Protocol terms this is Scope 1, Scope 2, and Scope 3 Upstream.

Emission factors

Depending on the type of merchant, we use either an Environmentally Extended Multi-Regional Input-Output (EE-MRIO) model or a direct spend-based method. Each has its own strengths and data sources.

Environmentally Extended Multi-Regional Input-Output (EE-MRIO)

EE-MRIO modelling integrates environmental data with traditional input-output economic models. An economic input-output model maps the relationships between industries and sectors within an economy. The Multi-Regional part refers to the fact that this model includes multiple regions, factoring in the imports and exports of each industry in each region. By extending these models with environmental data, EE-MRIO allows the calculation of environmental impacts associated with economic activities, such as carbon emissions.

We use EE-MRIO because it provides a comprehensive view of carbon footprints. It traces emissions through complex supply chains, enabling identification of key sectors and activities responsible for significant emissions. This is especially useful for goods and services with complex supply chains.

For example, an EE-MRIO model for an electric car manufacturer accounts for emissions not just from manufacturing, but also from raw-material extraction and processing, component production, and the transportation of materials and components to the assembly plant — covering all cradle-to-gate emissions (Scope 3 Upstream). We then add direct emissions from product use where applicable.

Direct spend

Direct spend-based estimates the amount of a commodity purchased from the spend, then uses an activity-based emission factor for that specific commodity. We source emission factors from official sources, aligned with government and other reporting guidance. This method is useful when the commodity is simple and has a high impact: gas, electricity, and fuel are the most common examples.

For example, for fuel purchases at petrol stations in the UK, the factor is derived from fuel price data and DEFRA activity-based emission factors: we estimate how much fuel was purchased, then estimate emissions from that volume using the official DEFRA conversion factors.

Further adjustments

We make several adjustments to increase the accuracy of spend-based estimates: price adjustments, additional user information, and weighted category mappings.

Price adjustments

When purchasing an item, part of the spend goes on production, part on taxes, and part into profit and transport margins. These need to be adjusted for, as they have different emission factors (nearly 0) and would otherwise affect the overall calculation.

Emission factors are for a specific date, so we need to account for inflation since the publishing of a database. For instance, if 1 litre of petrol cost £1.20 in 2022 but costs £1.35 in 2024, we adjust the emission factor in line with the new cost. We use monthly published inflation data to account for rapid fluctuations. In the UK in 2022, both electricity and gas prices rose by more than 30% over just one month — capturing price fluctuations monthly is important.

All of these adjustments are unique to the country or region of the transaction.

User habits

By gaining an understanding of the user behind the transaction, we can change some assumptions in the emission-factor calculation. A simple example is diet: a vegetarian will, on average, have fewer meat products in their basket than a meat eater, so we adjust the emission factor accordingly.

Weighted category mappings

Emission factors are created for individual products, not individual merchant categories. We take a weighted basket-of-goods approach — for example, a single groceries factor is built from a selection of grocery products based on regional consumption habits.

This means the models can be used with merchant-categorised data (which is the case for most bank transactions, typically using Merchant Category Codes or another merchant-oriented categorisation). It also enables us to map factors to custom categorisation schemas, provided the schema has enough granularity to accurately categorise transactions from an emissions perspective.

Intelligent categorisation

To improve the quality of data input to our models, we developed intelligent categorisation models that use machine learning to categorise individual transactions with human-expert-level accuracy. The benefits:

• Increased coverage. For card transactions a category is usually available (e.g. MCC), but for bank transfers and direct debits there's often no category, which means emissions can't normally be calculated. This is a big chunk of total spending for SMEs and consumers, particularly on high-emitting categories like energy utilities. Using intelligent categorisation increases the fraction of transactions we can estimate. For UK SMEs, this raised overall coverage from 60% to 91% of transactions.
• Increased granularity. Transaction category schemas aren't designed for carbon accounting, but transaction descriptions often contain more information than the category captures. Utilities is a good example: a water bill versus a gas/electric bill can be an 8× difference in emissions factor, yet this distinction is often missing from transaction schemas. Intelligent categorisation reduces the need to make assumptions, increasing accuracy.

Data sources

The following is a non-exhaustive list of data sources we use. Specific countries or regions may use different sources.

Validating estimate accuracy

We use several methods to test the accuracy of our methodology:

• Benchmarking against third-party sources (noting that differences can be due to varying methodologies rather than errors).
• Evaluating emissions profiles of different personas and comparing results to official and peer-reviewed data.
• Assessing model outputs and reviewing outliers to ensure the reliability of emissions factors.
• Assessing the uncertainty introduced by any custom categorisation schema via a full data assessment with real user data before going live.
• Reviewing methodology and code changes with external partners.
• Comparing per-capita emissions calculations with external reports to verify close agreement.

For more information, please get in touch via our website.

Partnerships

Our carbon data team continually refines our calculation methodology. We work with experts including members of The International Reference Center for Life Cycle Assessment and Sustainable Transition (CIRAIG), academics from the University of Bath, and researchers from the Basque Centre for Climate Change, to review all significant changes to our methodology. We also frequently consult experts in the environmental-impact measurement field to ensure we stay abreast of the latest developments.

We also work with experts on carbon reporting frameworks, such as sustainability consultants from OnePointFive, to ensure our calculations are fully aligned with GHG Protocol guidance and other frameworks and standards.