The decision to build Ada was taken in early 2024. The purpose was to improve the low quality data levels present in our client’s reporting, for both purchased goods and services, as well as investment impacts. We noticed that the reporting quality was too low for genuinely actionable insights and not granular enough to support procurement or investment choices that aligned with carbon reduction plans and execution.
Ada is built off the expertise of our team who first devised a Scope 3 tool back in 2006. The purpose of this tool was to provide accurate real world impacts for aviation emissions, surpassing the simple averages used at the time for long haul, short haul and domestic flights, to deliver accurate results, by airline, aircraft and route, thereby helping to inform better purchasing decisions and also support emissions reduction through purchase choices. Building this required extensive airline fleet, aircraft capacity and loadings, supported by academic research on engine performance. As such, when seeking to tackle the broader Scope 3 issues that Ada needed to solve, we were under no illusions that this would be a quick development and deployment.
So it has proved for Ada, but sometimes the end result is worth the effort to achieve the aim of delivering the highest quality. We are not in pursuit of the highest quality simply as an academic endpoint, but addressing how poor quality data makes company reporting very unreliable and lacking in commercial utility, especially when trying to meet targets.
The start point for our research and development, was to harness all the publicly available corporate reporting and assess its quality and accuracy. In tandem, we reviewed other models, like EEIO, that are used by many large organisations for reporting to identify how accurate these frameworks are when compared to actual corporate reporting.
This left us with the challenge of how to solve the wider problem of assessing Scope 3 impacts, where the vast majority of companies don’t report anything. Reporting by companies is still a very small percentage of total registered companies globally, mostly they are larger companies who are required to report, or who have voluntarily chosen to in some cases. To address this problem, we undertook research into company financial reporting and common data points to assess their correlation or lack thereof, with emissions.
After we had reviewed dozens of financial and demographic data points we discovered several that had statistically significant correlations with reported emissions. Expanding the control set confirmed the validity of these findings. This allowed us to use an LLM to provide predictive analysis for non reporting companies to an accuracy rate of over 90%.
Additionally, by using the widest possible data set to build Ada, we could begin to get more granular within existing reporting categories, providing better insight to companies, including sector averages and geographic specificity.
The key strand through all of this work, was the importance of the collected data and its verification. Poor base data will always deliver poor outputs, as we have seen in existing framework models. We used analysts with carbon accounting expertise to extract the data, verify and unpick inconsistencies, and assess the reported quality. This is a slow process and hard to automate accurately as we discovered. It is compounded by the lack of any single accessible database that holds corporate emissions data, even in the UK and EU where compliance reporting has been mandated for several years.
AI seemed like a logical workflow solution to address the speed of quality data collection and front end automation of sectors and corporate matching, which users often omit from their uploaded data sets. As we discovered, using experienced AI experts, whilst AI has some great workflow potential and real world impact, carbon emissions reporting is currently not one of them. Data collection accuracy was forecast at c90% but once checked by our analysts, this accuracy rate was found to be between 20-30% depending upon geography. We worked closely with our AI team to try and find ways to improve accuracy and enable the automated real time collection of published data for individual companies, but were ultimately unsuccessful.
There are some key reasons why this was our outcome from AI, and why we have moved back to using data analysts for primary collection and validation. Firstly, the collection of data that is non standard and requires evaluation, supported by carbon accounting expertise. This, as yet, could not be replicated. Non standard reporting includes bespoke, company specific metrics, merging of location and market based emissions, wildly differing report formats and content, location of published reported data (website, investor report, dedicated sustainability report etc). The list is extensive.
The UK has its own singular reporting anomaly in Companies House, where large companies are required to report their emissions. They all choose to do so to a greater or limited extent that is non standard. Reviewing this data from public filings is complicated by the fact that these reports are converted into a pdf format, that is resistant to OCR analysis as a primary locator of data. The accompanying API is not set up for emissions returns.
Whilst engaging with our fellow professionals in the Carbon Accounting community, their main goal for a tool like ADA was to use multi year data and insights. This would improve utility outside of initial reporting and support better choices. Critically, it would allow those Carbon Accounting professionals to easily show clients YOY changes which psychologically helps keeps organisations engaged by gamifying their emissions reduction. This requires harvesting multi year data that has to likewise be analysed and integrated, but which we knew would meet our core objective of utility alongside high accuracy. We have made great progress on this and will incorporate a new set of features for Ada due for launch in January 2026. Our new and expanded database and outputs will be available during September 2026.
Outside of the challenge of building Ada, we have learnt a lot along the way that is not entirely encouraging for the state of carbon reporting or the ability of companies to manage emissions from their reporting. Here are a few.
- Many companies have targets but not all companies with targets actually report emissions publicly on an annual basis. A large sample of several thousand companies with SBTi targets revealed that for many there were no public reports. A small fraction were hidden behind limited access firewalls or investor only access. It calls into question the transparency and actual progress that those with short term 2030 targets are making.
- A major obstacle to current reporting is the use of outdated data sets to drive results, sometimes 6 years old. This provides a number for annual reports, but not one that is reflective of recent trends in reduction from reduced input on emissions, electricity being a notable example where increased deployment of renewables is reducing emissions per kwh.
- The frequent use of EEIO (Environmental Economic Input Output) models was first introduced over 15 years ago to provide an emissions impact by product type, for example a soya bean. This reporting was upgraded for several years and then discontinued. There are inherent problems with using a model that implies homogeneity of price and process and clearly not all soya beans are produced to the same level of efficiency in every producing area. During a hiatus in which EEIO factors were not reported for several years, a number of carbon reporting tools and companies began to use the last reported factors and overlay these with inflation as a way to update the factors. For companies who started reporting as inflation surged from a relatively low and predictable range, emissions increased and then fell as inflation globally peaked. When UK impacts were assessed against the overall UK emissions outputs company reporting was rising at a pace that was not compatible with the reductions seen in the national inventory. When launching the Beta version of Ada we analysed a selection of companies at random with specific products and then used EEIO factors to compare results. The inaccuracy of EEIO as an emissions factor was in the range of 2000% from actual emissions.
- Broad sector averages are not helpful to accurate reporting and these are often present as the primary approach to Scope 3 reporting. In Purchased Goods & Services they have little utility in enabling emissions reduction planning and are not reliable impact markers when, as often, they are built on outdated data sets and are of course unable to distinguish between carbon efficient and inefficient suppliers. This degrades the ability of procurement teams to work with ESG teams to act effectively in addressing emissions reduction.
- Finally the quality of reported data is extremely variable, irrespective of the size of a company. There are examples of outstanding corporate reporting but these are unfortunately not the majority. All our data is assessed for data quality using the PCAF approach as its guide. Only a minority of companies gain a data quality score of 1, compatible with undertaking limited or reasonable assurance of their reporting. Worryingly some companies make this claim, but our analysts are able to identify where these claims are unsupported by known standards e.g ISO14064-3. Some companies claim to have marked their own work for assurance, a contradiction of the process of external assurance, whilst others claim assurance from standards that are a verification, not audit or assurance standard as claimed. In short there is a lack of understanding and factual accuracy in data quality reporting.
Ada has been built to address these challenges and provide direct subscription feeds for instant online reporting or via a dedicated API. Our bespoke analysis can provide emissions guidance on procurement, asset benchmarking and a range of specific analyses using our extensive data sets. This will be a continuing challenge for Ada that is managed through the full time use of data analysts to improve and increase data quantity, quality and accompanying insights. With over 70% of emissions coming from Scope 3 for many companies, intelligent insights will be the main route emissions reduction. The alternative is either doing/buying less or reporting poor quality data whilst hoping for emissions reduction benefits. Neither is a commercially desirable outcome and the Ada team will continue to meet the challenge of commercial and emissions goals for our clients now and in the years ahead.
