In a world where sustainability and carbon emissions tracking are becoming increasingly important, businesses need to make sure they're relying on the most accurate and up-to-date data available, but some aspects of indirect - Scope 3 - carbon emissions can be difficult to calculate.
Typically, the greatest challenge is in the supply chain – starting with purchased goods and services – where suppliers are not themselves geared-up for carbon foot printing.
We’ve seen some interesting and some pointless applications of ChatGPT from OpenAI recently. However, with ChatGPT’s level of general knowledge, this problem is starting to get easier, meaning organisations can begin to understand and tackle key sources of emissions in their supply chains.
Where do most carbon emissions come from?
When it comes to business carbon emissions, the largest proportion comes from Scope 3. This is because Scopes 1 and 2 deal with activities that are specific to that part of the value chain where a company operates - whereas Scope 3 relates to carbon emissions resulting from activities for which it is indirectly responsible - up and down its supply chain.
When it comes to Scope 3, for most organisations the largest category is Purchased Goods and Services. However, it is one of the most challenging to account for, report, and monitor.
Why should you take a granular approach to Scope 3 emissions?
Traditionally, accounting for Purchased Goods and Services is done without using invoices – either by categorising vendors - or using a crude measure such as cost centre. An OpenAI-powered approach maps each purchase to a specific industry standard category and from there a ‘spend based’ estimation of carbon footprint can be defined.
Combining a granular approach with OpenAI APIs gives firms the ability to proactively uncover risks and opportunities from within their supply chains. It also reveals where you can reduce costs whilst still reaching your sustainability targets - by enabling enhanced procurement analytics.
Categorising products per purchase also makes it easier to compare emissions across different purchases and product categories. This can help you to identify trends, opportunities for emission reduction, and areas where emissions are particularly high. When you group spending by vendor, you may lose sight of these detailed insights and make it difficult to identify areas for improvement.
This granular approach enables you to engage with vendors more effectively. By understanding the emissions associated with different goods and services, you can work with your vendors to identify opportunities for emission reduction and encourage the adoption of more sustainable practices.
What are the drawbacks of grouping Scope 3 emissions by vendor?
When you group spending by vendor and apply a single emission factor to all products from that vendor, you're assuming that all purchased goods and services have the same emission factor. This can be inaccurate, as goods and services from the same vendor can have different emissions levels, manufacturing processes, and supply chains, which yield different emission factors.
Take, for example, technology companies that offer professional or consultancy services as well as hardware like laptops, monitors, and phones. This is the same vendor, but these two sides of the business result in two completely different emission factors.
It’s important to say that no matter how accurate the categorisation, the spend based approach also has drawbacks and for the highest emitting categories / vendors, actual emissions (‘activity data’) should be substituted.
How to calculate Scope 3 emissions per purchase
For smaller organisations, going through product descriptions and vendors manually to classify purchases might be the best way to proceed. But what about companies that have thousands of vendors and hundreds of thousands of purchases? The exercise may be impractical from the outset – but even if that does not prove a blocker, it needs to be repeated every year - unless you can adopt an approach that combines speed and accuracy using the power of machine learning.
One of the fundamental steps to combining per purchase and per vendor emissions data is the use of Natural Language Processing (NLP) models to classify this type of data for different types of businesses.
As long-time innovators with this technology, we've developed our own in-house model to perform this task, but it still needed some labelled data, which again requires human effort and time.
The release of OpenAI APIs allowed us to perform this task seamlessly. Before the release of GPT–4, GPT-3 was one of the largest models available, with a capacity of 175 billion parameters. These large language models (LLMs) are trained on a massive corpus of data with a diverse range of sources, including books, articles, websites, and other text-based resources.
This training gives LLMs a broad base of knowledge from a variety of topics, including science, technology, politics, history, and culture. LLMs therefore have the ability to classify purchased goods and services using the industry standard classification systems codes - whether that be NACE, SITC, NAICS, HS, or something else.
If you want to achieve state-of-the-art performance on the categorisation of Purchased Goods and Services, all you need – aside from the data, of course – is the recent text-davinci-003 update from OpenAI, some parameter tuning, and prompt engineering.
GPT-3, specifically the variant based on the text-davinci-003 update, is one of the most advanced and powerful models for natural language processing classification tasks. So, it's perfect if you're looking for an accurate, fast way to classify your Scope 3 reporting. At NTT DATA we are using OpenAI on Azure, which carries strong data privacy assurances.
How can I incorporate AI into my Scope 3 emissions reporting?
Many firms have set net zero targets for 2030 or 2040 – these deadlines are now looming. Legislation is becoming stricter, along with reporting audits, lobbying, and investor priorities shifting towards sustainability and carbon neutrality. The direction of travel is ever greater detail and accuracy.
At NTT DATA UK&I, we're continually evolving our own solution to this problem, the Single View of Carbon, to address the data challenges businesses might face when reporting their carbon emissions. Adopting a granular approach and leveraging the power of advanced AI technologies to calculate carbon emissions from purchased goods and services is critical.
With Single View of Carbon, businesses can now easily and precisely measure their carbon footprint to make more responsible decisions. By implementing OpenAI APIs as part of this unique solution, you don't even need to manually collect data or optimise your models by hand.
If you'd like to know how we can help you and your organisation optimise your sustainability reporting without compromising time or resources, get in touch today.