GoClimate Uses AI to Slash Carbon Reporting Time and Boost Action

Climate action is progressing too slowly, and companies are getting stuck in time-consuming administration instead of reducing emissions. GoClimate now uses language models to automate the most extensive parts of emissions calculations directly from bookkeeping data. The result: radically faster analysis, lower costs, and more time for real action.

GoClimate is taking the next step in digitalizing climate work, using artificial intelligence as a turbo boost. By leveraging AI to analyze accounting data, the company can perform sustainability calculations both faster and smarter—without compromising human expertise.

"This might be the most powerful technology ever invented. And if we are going to use it for anything, it’s to make the world a better place," says GoClimate’s new machine learning engineer, Leo Hiselius, who is responsible for developing efficiency and ensuring accuracy in AI work.

Previously, collecting and interpreting data from invoices and receipts required many hours of manual work. With the help of language models, GoClimate can now automatically identify products, quantities, and prices and link them to the correct emission factors. The result is fast, accurate climate calculations based on companies’ accounting records.

"With language models, we don’t have to write rules for each invoice. Simply put, we can say: ‘Here’s the invoice—give us the products, quantities, and prices.’ And the model handles the task, no matter how the invoice looks," explains Leo Hiselius.

To write all the manual rules, GoClimate would have needed hundreds of developers coding—something that wouldn’t have been sustainable given the company’s growth and new clients. However, the work is always supervised by human experts.

"Language models are fantastic general problem-solvers. The core of my work is keeping the accuracy as high as possible," says Leo.

How GoClimate Handles Ethical and Environmental Challenges with AI

AI consumes a lot of energy, especially during training. But recently, models have become much more energy-efficient. For example, Google’s new Gemini models use 97% less energy per query than a year ago, and each AI query corresponds to the energy used by watching TV for three seconds.

"AI requires energy, but the gains in efficiency and climate impact are enormous when we can help thousands of companies understand and reduce their emissions," says Leo Hiselius.

GoClimate only uses language models hosted on servers within the EU to ensure that customer data never leaves the EU. Additionally, GoClimate uses services from servers where the majority of energy comes from renewable sources. Most of the servers are in Finland, where 98% of energy is renewable, but some resources are also in Belgium, where 84% of energy comes from renewables.

Looking ahead, GoClimate is preparing for the next step: multimodal models that can interpret both text and images—and in the future provide companies with customized advice to reduce their emissions.