Methodology

The primary basis for all carbon accounting is the Greenhouse Gas Protocol (GHGP) Corporate Accounting Standards (CAS). The standard defines how to account for emissions and has been used as the basis for sector frameworks. In relation to Scope 3 the GHGP CAS allows for some degree of estimation within reporting where data is neither accessible and/or too expensive for a company to reasonably obtain. It recommends approved conversion factors for activity data. These approved conversion standards, such as BEIS/DEFRA in the UK or the EPA in the US, provide conversions with clear methodology that generate high-level estimates of emissions from activities. Granularity in these factors varies considerably from real world emissions.

One of the sector standards for carbon accounting is the Principles for Carbon Accounting in Finance (PCAF) which includes a scoring system from 1-5 with 1 being the best quality data, usually supported by limited assurance, and 5 being the lowest quality, often based upon sectoral averages and estimated activity data. Ada uses PCAF as the basis for its data scoring evaluation and replays these in its outputs, allowing clients to visualise increasing data quality over time.

The team who created Ada have extensive experience in audit quality carbon accounting and ISO limited assurance. The work they undertake regularly for clients across multiple sectors and geographies has provided a depth of insight and experience that informs the company’s work and the build of Ada. The approach taken is conservative leaning, in pursuit of reduced risk and optimised outcomes for multiple use cases from reporting to decarbonisation pathways.

The primary emissions data that Ada uses is Scope 1 & 2 company-level emissions, from direct energy usage and purchased electricity, heat and steam. This data can include both market-based and location-based reporting. The variability of market-based reporting, absent limited assurance, means that it can be used, but attracts a lower data quality score. The key variable here is the Scope 2 Value Chain Criteria published by GHGP and regularly updated. These criteria specify which types of “renewable energy” can be zero rated and which should not be.

Most companies with greater than £36 million or EU currency equivalent are required to report emissions for Scope 1 & 2 at a minimum. Scope 3 has significant optionality and associated quality issues. It is not used in Ada calculations, both to avoid double counting of emissions and to ensure a comparable analytical framework for all company records.

Ada’s data is collected from thousands of companies whose reporting is available in the public domain. In addition to emissions data, revenue, FTE and selected key financial indicators (KFIs) are collected.

The data quality delivered by Ada has several checkpoints to ensure that it is able to deliver high-quality outputs. All reported company data collected by Ada analysts is assessed for accuracy and completeness. The system is designed to flag company results that may be incomplete, anomalous or contradictory. These companies are segregated for further review by our data and accounting specialists, prior to being introduced into the model. Additionally, the team review core data, output results and key data points monthly, to ensure that the model and its associated processing capability delivers the highest standards of accuracy. External specialists are engaged to review the model function and data ingestion, to ensure that it is maximising the limits of machine learning without compromising data quality.

Ada uses weighted sector averages to deliver a more accurate estimate of companies where no public reporting is evident or reported in multiple sources that Ada has scoured. Weighted sector averages reflect all the reported companies in a given sector to ensure that dominant companies in a sector do not over estimate averages, even using FTE or Revenue as balancing factors.

The use of broad sector averages of multiple types was assessed in the research programme for Ada to evaluate the impacts of commonly used sector inputs, including inflation adjusted input/output models where conversion factors are over a decade old, and broad spend based models using primary intensity metrics only.

Ada uses a machine learning large language model that is unique and proprietary to the product and has been built based upon 2 years of research into company data, key financial indicators (KFIs) and their relationship to reported emissions. The research programme reviewed over 20 KFIs that are often used in financial reports to establish statistically valid correlations or reject non correlating patterns between KFIs and financial indicators, as well as an ability to eliminate outlier data that might skew overall results.

Ada delivers company level estimates that are up to 90% accurate when measured against a control model of actual company data. This in turn is able to strengthen the sectoral averages where unknown company values may be present.

Ada clearly states the match rate of companies submitted to us versus our database. The match rate is the number of companies for which a direct match exists in the system for Scope 1 & 2 reported emissions. Unlike other products, the match rate is clearly reported on Ada’s dashboard so that a user is able to determine the extent to which sectoral averaging may influence their reporting. This allows for increased accuracy in reporting in line with recommended accounting treatments. The match rate also enables a fully differentiated data quality score that is most commonly used in financial services reporting of investments and the ability for clients to see changes/improvements in data quality over time.

When Ada processes incoming client data it will seek to verify the correct company name as a primary screen of the submitted data points. Ada will reject and clearly report to users the company records that cannot be processed and the reason why. This is a common problem in company matching where the company name supplied does not exactly match a legal corporate entity, either due to an abbreviation or the company being a group subsidiary. Where supplier data is provided with a VAT number, Ada can verify the exact company name match from registry records to facilitate matching of correct companies.

All sector average data is an estimation. Ada produces very high accuracy in estimated data, when compared against company data sets where reporting is known. Control sets are used to test the relative accuracy of company and sector estimations. To further support a full understanding of the data, Ada produces a predictive variance analysis to provide users with an understanding of how much the data results may deviate from real world emissions. This analysis is linked to the wider data set and the data scoring. Data scoring methodology is based upon PCAF scoring and Ada is able to make an informed view on how much differential may exist from the estimated impact. This has particular relevance to the use of Scope 3 data in baselines and target pathways to assess potential variance in outcomes.

Ada’s machine learning model is not directly linked to its front end reporting module. This ensures that results do not change on a daily or even hourly basis throughout the year as Ada acquires more data and refines its own processes. To streamline data flow, Ada is updated every three months with the latest core data sets in flat files that are reviewed by the Carbon Responsible team prior to upload. Users are notified of updates and can rerun their data sets for the latest insights into their impact and accompanying results. As more companies report and existing reporting improves so will the results from Ada.