Sustainability Reports: AI analyzes topics and supply chainsDoing the Research series

15 September 2025, by Newsroom editorial office

Photo: UHH/Cordes

Sustainability topics in the Lufthansa annual report for 2024, depicted in a “topic model.”

When corporations address sustainability they do it for marketing purposes and often in standardized ways. Hannes Cordes, who is working for the professorship for logistics and supply chain management headed by Prof. Dr. Guido Voigt, is looking at how corporate sustainability reports can be used to create greater transparency with regard to supply chains.

The research team is looking at how AI language models can be used in conjunction with sustainable supply chains. What is the focus of your project?

Whoever wants to know about a company’s sustainability can look at its sustainability or business report. However, what if we want to know less about the company itself and more about its suppliers? And what about the suppliers of the suppliers? Complex networks of relationships arise quickly, creating a seemingly endless flood of information.

In our research project, which I am writing my dissertation on, we are looking at how we use AI-based methods in natural language processing, meaning computer-supported understanding of human language, to illuminate this flood of information and how we can use these for research on sustainable supply chains.

How have sustainable supply chains been analyzed so far?

To date, corporate evaluations from external providers have been used to analyze complex supply chains or networks. To some extent, these are not transparent and they vary greatly from provider to provider. Interviews are an alternative to this, or surveys in cooperation with the company. These can yield a high degree of detail but they are time- and cost-intensive.

This is where our research comes in. Using natural language processing to evaluate sustainability and business reports, we can quickly gain systematic, detailed information to scale. The goal is to model networks in which we have transparent and comprehensive information about what and how particular sustainability topics end up in reports.

Which methods concretely are you using?

I am looking primarily at sustainability reports that use “topic modeling,” meaning reports that cluster topics. The way this works is that a language model, a so-called large language model, translates text data into numerical values. These values can then be positioned in a multi-dimensional space and the distances between the points can be measured. The results are presented so that people can interpret them. This means that it is clear to me whether a certain topic, for example the CO₂ balance, appears in the report. This makes it easy to see what is being reported.

But we are also using algorithms to classify, for example, to identify positive, negative, or neutral sentences or to categorize different sustainability dimensions (environment, society, ethics). This generates a better idea of how reporting is done. With regard to methodology, we worked closely with Prof. Dr. Anne Lauscher. As a professor of data science at the University of Hamburg, she helped us along with her expertise in natural language processing.

So then you have initial data for a company. How do you get to the information about the supply chain?

Results are added as data points to a network model. In this model, the companies that we want to analyze are nodes. Their business relationships among one another are the connections. The focus might be an industry but also a company with its suppliers and customers. The evaluation follows as a social network analysis. When we do this, we see companies as social organizations whose activities we depict in similar ways and can analyze as we would human social structures.

For example, we can find groups of companies within a network that work closely together on certain topics. Or bridging actors that do not have many connections to one another but transport topics from one group to another. Or little islands that, isolated from the larger network, advance their own issues. And the time component is very interesting. This allows us to understand the spread of individual sustainability topics to the network.

Do you have a concrete example?

I am looking concretely at airlines that handle freight for logistics suppliers. This is a little-studied set-up because logistics suppliers rarely turn up in research as customers. Furthermore, service providers in supply chain management generally command less attention that manufacturing companies. And yet they play an important secondary role, especially in the organization of supply chains.

Based on the analysis, I was able, for example, to look at how they handle synthetic fuels, also known as “sustainability aviation fuels.” These are an essential building block in aviation’s decarbonization strategies. The long-term goal is to reduce about two-thirds of all CO₂ emissions in the industry. On the part of airlines, the issue has become much more important in the last several years. However, when it comes to the customers, meaning the logistics suppliers, reports largely ignore this.

So it seems in this case as if the customers have little interest in the measures that the providers are taking. And yet airline emissions also show up in the CO₂ report, just in another place. They would also benefit from increasing use of sustainable aviation fuels. The next step would be to look at how fuel producers, airport operators, airplane engineers, and senders in other industries report on these matters.

How do you put your findings into practice?

I present the methods and results at conferences and in talks. At events in Hamburg especially it is easy to make contact with representatives in aviation and logistics, for example, from Airbus or Lufthansa Technik.

One thing that gets discussed a lot is how the reported activities do not necessarily correspond to actual activities. For example, there might be a statement of intent that is later just dropped or postponed. And the topic of green-washing always pops up. These are all good and legitimate questions that, sadly, lie outside the scope of my work. For me, however, these issues are a very good starting point for further research. As soon as the corporate perspective is completely modeled, it can be compared to a host of other data and evaluated. For example, with reports from the media or environmental protection organizations. Then we can precisely uncover any discrepancies.