Traceability from the bottom up
GIST Impact is turning to AI to help collect, validate and analyse sustainability data – but traceability, transparency and explainability are key, says Mahima Sukhdev
Environmental Finance: How is AI changing how GIST Impact collects and analyses sustainability data?
Mahima Sukhdev: Nature impacts are very location-specific, so where a company’s assets are located really matters. But we’ve found around one-third of assets in third-party datasets are mistagged: we discovered a drive-through coffee shop tagged as a coal mine, for example. So we’ve used computer vision along with multiple other location- and company-specific features to train our AI models to verify asset locations at a global scale while demonstrating their reasoning: we’re able to validate, with confidence, close to 4 million asset locations for just over 10,000 companies. This is getting better every day. And we keep humans in the loop to ensure every insight remains explainable.
This is helping us build traceability from day one: so auditors, for example, can trace back each asset tag to the evidence; and so our clients can make confident decisions when it comes to nature- related data. Clean asset location data is a bedrock for everything else we build on top, whether it’s nature or climate risk, materiality assessments or impact calculations.
Our overall aim is to build a living ‘digital twin’ for every company. This twin incorporates data the company is disclosing, alongside that from other sources – trade registries, shipping bills, NGO reports, etc. – so we can create a detailed picture of the company’s impacts and dependencies and enhance it over time.We have multi-agentic AI models that take in new company disclosures and other third-party information and stitch it together in ways that are valuable for our clients.
EF: You have raised concerns in the past about the accuracy of data estimations: how do you approach private market assets or where disclosures are limited?
MS: Estimations are a necessary evil, because not all companies disclose all material nature and climate data. But the most common approach, using industry averages, is rather simplistic, and the use of estimations is often not clearly disclosed. This is important to address, as some big decisions are made using this data.
As a technology-native company, we have developed sophisticated models to enable us to gap-fill missing environmental data, which we train on a very high quality and granular underlying corporate feature set.
We’re getting fantastic results: as an example, our carbon estimation model is at above 90% R^2 (an industry-standard measure of accuracy): a recent study by one of our clients found the market standard for carbon estimation was only about 60%.
EF: What particular challenges do banks face calculating climate and nature risk in their loan books?
MS: Banks have vast loan portfolios, and they urgently need to understand how climate and nature will impact those loans. But there are many challenges associated with that. First, they tend to have a large geographical blind spot, e.g. using the company’s headquarters as a proxy for their location, thus failing to capture the actual risk to its operations. Second, they usually overlook supply chain risks, due to a lack of understanding of the dependencies a company might be subject to up its supply chain – it’s quite a data-scarce area. We’re using AI to help piece that together.
Third, climate and nature risks have often been siloed by banks in the compliance function, partly because of the challenges in converting the data into financially meaningful metrics.We’ve solved for that by integrating over 30 physical, transition and supply chain risks into the same risk models that banks use for default and credit risk probabilities.
EF: GIST Impact has recently teamed up with the Natural History Museum. What was the thinking behind the collaboration?
MS: The Natural History Museum has developed its Biodiversity Intactness Index, which is the global gold standard for measuring ecosystem health, built over many years of research. By deploying the Museum’s world-class science along with GIST Impact’s world-class corporate and asset data, and adding our AI-led explainability and contextualisation, we provide investors with frontier and forward-looking insights on nature-related risks and opportunities.
We have also joined forces with other mission-driven non-profits like Global Canopy and IBAT. We don’t believe in black boxes or reinventing the wheel: the climate and nature crises demand openness and speed. We like to team up with these organisations, who are mission-driven like us, and convert the nature-related data they have painstakingly collected into actionable, investor-grade intelligence. This gets critical insights into the hands of decision-makers at financial institutions in weeks instead of years.
Mahima Sukhdev is chief growth officer at GIST Impact.
For more information, see: www.gistimpact.com/impact- intelligence/#climateNature
