[AI for Business 3] Case study: Using AI for sustainability reporting
How Groupe Bruxelles Lambert initiated the use of AI and implemented its first AI use cases
In one of the previous episodes (A Roadmap for AI Implementation), we explored a three-step process that businesses can use to get started with AI. Today, we will meet Francois Perrin and learn how he pioneered the use of AI in his company. Francois leads the sustainability activities of the investment holding Groupe Bruxelles Lambert (GBL), a client of my company Equintel. GBL is the 5th largest company in Belgium, holding major shares in companies such as Adidas, SGS, and Umicore with a total portfolio value of 17.6 billion EUR.
I will first describe the AI use case we implemented with Francois. Then, we will dive into an interview about his journey with AI.
AI use case: augmenting sustainability reporting
In our cooperation with GBL, we used AI to automate materiality assessments for sustainability reporting. Let’s decipher the use case.
Larger context
The role of sustainability is changing - for most companies, sustainability has been transformed from a matter of goodwill into a regulatory obligation. For example, in the EU, the new CSRD regulation forces companies to report on their sustainability efforts. This means that whole teams get bogged down in the details of reporting for months, not having the time and space for more meaningful work.
Scoping down the use case
Reporting is a complex process that normally extends over months and involves not only the sustainability team, but also other departments such as HR, Compliance, and R&D. To narrow things down and define a realistic use case for AI, we first break down the process into its main steps:
In the materiality assessment, the company defines the sustainability topics that will be included in the report. For example, a company in the mining industry needs to pay special attention to environmental topics and working conditions in its mines, while a pharma company needs to address topics like patient safety and access to medicine. The materiality assessment allows companies to “scope down” the content of the sustainability report. In the second step, the sustainability team collects data about the material topics. This often involves back-and-forth with multiple departments - they can address HR about topics like working conditions, R&D about patient safety, etc. Finally, with all the data collected, the team can sit down and compose the final report.
After analysing this process, we decided to start our development with the materiality assessment - mainly for two reasons:
It can be done using external data, such as media data, information from competitors, etc. Starting with publicly accessible data is a tremendous advantage for AI development. It allows you to kickstart the development without creating dependencies and getting into a conundrum of discussions about data privacy.
It is the first step in the reporting process. Simple, right?
Implementing the AI solution
We designed an AI system that basically performs the work of humans who collect external data about the industry and the peers and extract the relevant themes and topics. The outcomes are presented in a dashboard that can be customised with client-specific analyses. It also provides convenient functions for data export so users can continue working with the data on their own (remember - so far, we are talking about augmenting the user, not completely automating the task).
Capturing the impact
The impact of the solution is two-fold:
Increased efficiency: This is a no-brainer - you get the same result at a (significantly) lower cost. In the manual scenario, the materiality assessment takes months, and many companies also commission costly external services from consultancies like the Big Four. With AI, we were able to drastically speed up this process, delivering a first draft of the materiality assessment after one or two days of customisation. A comparison with the outcomes of the “manual” process showed an 80-90% match in the results.
Qualitative advantages: These are more subtle benefits of using AI - they might need some time to manifest themselves, but ultimately, they allow you to do a better job. For example, automation allows us to process much larger data quantities than a human could do, thus increasing the coverage and objectivity of the analysis. This provides a better basis for the subsequent steps, as well as for auditors who expect an explicit and defensible methodology behind the process.
Putting things into perspective, as we are flooded with glossy reports about world-changing AI breakthroughs every other day, sustainability reporting might not exactly look like the sexiest AI application. But once you crack it, there is a lot of appeal to be found - you automate and speed up a mandatory chore, relieve people from the grind, and allow them to focus on the meaningful work of addressing global sustainability challenges.
If you want to learn more about the use case, check out this webinar. Otherwise, let’s join my co-founder, Timo Heroth, as he discusses GBL’s AI journey with Francois. As we go along, I will be highlighting some of the key learnings and best practices for using AI in real-life businesses.
GBL’s AI journey
Timo Heroth: Francois, thanks for taking the time for this interview. Let’s start with a broad picture. Could you please talk about the overall context in which you are using AI?
Francois Perrin: Of course. Unfortunately for the entire sustainability field today, we've been dragged into a rather inefficient and time-consuming reporting cycle. With the new CSRD regulation, more than 50,000 companies in Europe are burdened with detailed reporting requirements. However, as a Chief Sustainability Officer, you cannot just focus on reporting. This would mean failing at your fundamental mission of transforming your company into a more sustainable enterprise. So you really have to look beyond regulations to propose something that is a little bit more fun and sustainable than just implementing another reporting process. But you don’t have time for this because you are busy collecting data and writing reports. With all the new developments of AI over the past two years, there is fortunately an alternative - you can use AI to save time and cost and free up your “human” resources for more impactful tasks.
💡 AI learnings:
Regulations are a rock-solid opportunity for AI automation. They create an additional overhead which often conflicts with profitability and business interests. Thus, companies welcome AI support to comply with regulations.
By automating a part of regulatory tasks with AI, companies can not only save time and cost, but also free up sustainability talent for tasks that are more important and satisfying.
Read more on identifying AI opportunities here.
Framing an AI strategy
Timo Heroth: Given this context, how did you approach an AI solution for sustainability reporting?
Francois Perrin: In general, we see AI as one of the key structural trends shaping the future of each and every industry, be it finance industry, more conventional industrial activities, consumer-facing products and services, etc. In my opinion, AI is redefining the conditions for successful business operations in the future. So just take AI as a tool and think about what you can achieve with it in your transformation. If you onboard AI at an early stage of a significant transformation process, you will get the full benefits through the process but also after the process, since you will have a significant advantage in terms of time and technological acumen. And for me, this is the real long-term benefit of any AI solution in our day-to-day activity.
In a more short-term perspective, we definitely save cost and time using AI to perform a number of tasks that otherwise would be conducted by a number of our own people and external consultants. So it's really a question of time, cost, and efficiency of the operation you are running. For us, the low-hanging fruit was the first step in sustainability reporting - the materiality assessment. Here, we identify those topics that are material for a company and need to be included in its report.
💡 AI learnings:
Two things are certain about AI: it is here to stay, and it can create value for most businesses. How exactly the value creation will happen is up for businesses to find out. In future episodes, we will be learning about the methods and tools to conduct and structure this discovery process.
Companies that get started with AI early will have an advantage over their competitors in the long term. Based on a good internal understanding of AI as well as a stack of existing AI components, they will be able to approach more and more complex AI use cases.
Rather than fleshing out a complete AI strategy from the beginning, start with the low-hanging fruits in your business, and keep defining and refining your AI strategy over time. Thus, while understanding the long-term disruption coming from AI, Francois uses AI as a tool to optimize existing processes.
Read more on identifying AI use cases here.
Quality and trust are make-or-break criteria for AI adoption
Timo Heroth: Thanks, François. Many companies have tried to implement AI but were disappointed by the quality and reliability of current AI systems when they go into production. How do you perceive the quality in the daily usage of our data, and how does it compare to the manual analysis?
Francois Perrin: There is always a trade-off. I've been looking at a number of AI models for sustainability tasks for now five years, and I have to say that things are progressing extremely rapidly in the space. When evaluating an AI model, the first question is: “How will the output look like? Is it badly wrong and unusable or does the output look reasonable?” For me, it usually starts with the assessment of the coverage, incl. checking the companies I want to research, and the data sources being mobilised. Then, the next step is to find out how the model is working, why it may fail at a task, and how I can help fine-tuning it.
If this first sanity check succeeds, you can hope that you will get something good out of the AI. Just keep in mind that there is always a chance it might get things wrong and that you will most likely never reach 100% correct answers with an AI model. This is OK. The target is not to achieve perfect quality - that would be a misunderstanding of AI capabilities and the efficiency gain. Rather, you need to ensure a proper definition and quality of the analyses you create in order to achieve an 80%-20% solution, and to take responsibility for the rest yourself.
If you can find an AI model that is delivering on a regular, reliable basis an 80% performance, you can then tailor the right set of questions and analyses that address your needs. Over time, you can focus on these outputs and optimise them. So, the most important thing is to start and to work towards a reliable set of relevant outputs. Thus, with Equintel’s AI solution, we are currently achieving a 80%- 90% match via an AI-generated materiality assessment.
Let’s contrast this with our original workflow without AI. Repeating manual processes is a real pain in our industry. A double materiality assessment can typically take up to nine months. Every time you do it, you will always start over with a new set of data, rebuild your spreadsheet, rebuild all the questionnaires, surveys, etc. And you will try to compare things that may have changed. With AI, that aspect of the manual task is something where you tremendously gain in efficiency and make your job more fun.
💡 AI learnings:
When selecting an AI model or application, use a general “sanity check” to filter out flops - try some very simple requests and see if the outputs are right.
Check explainability - can you find information about how the model works, why it might fail, and how to follow up on failures? Read more on explainability: https://jannalipenkova.substack.com/i/144725409/explainability-cracking-the-black-box-of-ai
Calibrate your trust into the AI system - understand that errors can occur and will require you to intervene. Still, you should be confident that the Al system can deliver reliable value when compared with a manual process. Read more on calibrated trust: https://jannalipenkova.substack.com/i/144725409/calibrating-trust
If technically possible, refine the AI system over time, narrowing down the set of relevant outputs for your use case and optimizing their quality.
Reusing an AI model across multiple touchpoints in the business
Timo Heroth: We are very happy that you achieved such a decent accuracy with our AI models. Did you also build up enough trust for more involved use cases, like the direct engagement with your companies?
Francois Perrin: Indeed, there are multiple ways to use the outputs. You can either use them as part of your own materiality assessment to kickstart the process and to be more efficient in terms of discovering, identifying, and putting down a first draft of your sustainability matter focus list. The second option is about control, where you use the AI to verify and challenge an existing materiality assessment. So you really have the flexibility to use the AI across multiple touchpoints in the process. It's a tool, and the main question you have to ask yourself is: How can you plug it in into your existing toolbox to help you accelerate existing processes and free up your people?
💡 AI learning:
To boost efficiency even further, the same AI models can be reused across different but similar use cases.
Embarking on an AI journey
Timo Heroth: Francois, by now, we can call you an expert in using AI. We just discussed the clean and mature use case of sustainability reporting, and specifically the materiality assessment. Let’s travel back in time for a moment - your AI journey with us started more than two years ago and was not always as straight and clear-cut. Could you describe why, back then, you looked at AI in the first place, why Equintel’s approach triggered you, and which challenges and benefits you encountered when introducing AI at your company?
Francois Perrin: Indeed, that has been a fascinating journey. Initially, we wanted to use AI to upgrade our annual risk review that was based on a mature, but rather conventional methodology. Every year, we would spend three months at the end of the year to review major risks for every portfolio company. We did the analysis based on external data sources, brought that back to the investment team and sat down to discuss what are the key risks on which we will focus next year. And these risks would be the key driver for our engagement with the company.
That approach was fine, and it is what our most advanced competitors would typically do. But unfortunately, we were missing an important characteristic of sustainability risk, namely their inherently dynamic nature. Sustainability risks are everything but static - they can emerge rapidly out of an unmanaged and unstructured situation but they can disappear also very quickly. By contrast, the mainstream risk analysis uses a static framework and tries to “mold” real-life risks into a rigid structure, which is just not the reality of modern business environments.
Therefore, three years ago, we decided to strengthen our existing sustainability risk management by adding this dynamics to our risk mapping, and that's why we eventually approached Equintel. We asked Equintel to track the sustainability exposure of every company we own, as well as of their larger peer groups. This monitoring was done on a daily basis, and we started identifying a number of emerging ESG risks that would not be captured by the conventional sustainability risk model.
Then, there was the question of figuring out whether a new emerging risk is there to last or will quickly disappear because it was, in fact, a non-event. So we use Equintel to monitor the full set of risks for every portfolio company. If we identify an emerging risk with the potential to significantly impact one of our portfolio’s companies, we have the opportunity to revert to its Board of Directors or other relevant governance committees. Based on Equintel’s data, we are in a position to request the relevant set of information and feedback from the company’s management and decide on potential level of mitigation.
This is how we add dynamics to our sustainability risk assessments. In a number of occasions across our portfolio, Equintel gave us the very first early signal on sustainability matters that were developing in a significant way and became material ones two to three months later. We were in a position to leverage on this information edge to start addressing the issue via the different governance bodies of our portfolio companies and to ensure that risks were properly addressed.
Timo Heroth: Thanks a lot, Francois, for taking the time and sharing your insights with us.
💡 AI learnings:
Francois started his AI journey by seeing flaws in the established process of risk review. Traditionally, this process is static, but a market environment that is changing quickly requires a more dynamic approach.
AI injects dynamics into the process since it is able to analyse large data quantities on a daily basis.
By upgrading an established practice with AI, GBL could improve its overall performance and strengthen its competitive advantage.
Wrapping up
As a final thought, I would like to emphasise the importance of combining domain expertise and tech skills in any AI project in the B2B space. Through a tight collaboration with GBL, we were able to directly ingest Francois’ experience “from the field” into the solution. Timo, who is finalizing his PhD in sustainability, contributed ideas from the newest academic research. As we went on, both of them turned into AI experts, while I can now spell out dozens of the acronyms that are routinely flying around in the sustainability space.
That’s it for today. As always, please share this post with interested colleagues. Also, get in touch if you have feedback or questions, want to share your own insights on the discussed topics, or have another topic that you are curious to learn about in future episodes.
I wish you a great first summer weekend and look forward to sharing the next episodes!
Best wishes
Dr. Janna Lipenkova