Kelly Meijers about sustainable value creation with AI

What is sustainable value creation?

This question seems obvious. Sustainable value creation is of course about creating value, not only for now, but also for the long term. And there is more to it than many organizations realize, Kelly Meijers, sustainable AI strategy specialist at Conclusion AI360 explains. "You not only have to develop an AI model, you also have to manage it and continuously monitor whether it is still performing well. After all, if the environment changes, the model must change with it. You will also need to train your employees. There may be resistance among these employees to the use of AI, so you will have to pay a lot of attention to innovating their work. You will also have to adjust job profiles, because tasks will be performed differently. In addition, there are laws and regulations that you must comply with, such as the European AI Act that is now coming into force. And I haven't even mentioned the other meaning of sustainability: the impact on our living environment. Complex AI models consume power and a lot of water is needed to cool the servers. That's not exactly sustainable."

In many companies, the focus is often more on the short term than on the long term. This is a pitfall, especially when applying AI, Kelly says. "Because developing an AI model or even purchasing and using an off-the-shelve model is very easy. In addition, many companies pay particular attention to applications that appeal to the imagination, such as the use of large language models. They forget that these solutions are also very complex, and therefore draw more in the areas just mentioned. That quickly increases the chance of costs ultimately being higher than the revenues." 

What is the right way to go about it?

Sustainable value creation always starts with the organizational strategy, i.e. the company objectives and ambitions, Kelly explains. "What are the bottlenecks or hurdles that prevent objectives from being achieved and ambitions from being fulfilled? What data do we need to solve this problem? Only then do we start looking at which technologies we can use. This does not always have to be AI, sometimes a simple BI solution is just as good."

Responsible AI transformation blueprint

To ensure that all aspects receive the attention they need, Conclusion Intelligence has developed a Responsible AI transformation blueprint. As mentioned previously, this blueprint starts with strategy and objectives. Next, the AI Readiness of an organization is mapped out and, based on the Responsible AI Framework, it is recorded in detail how mature the organization is in terms of the six dimensions that play a role in sustainable value creation.
Kelly advises to always start small, with a problem that is current in your organization. "We are developing such a use case into a prototype, through which we demonstrate that AI can create value. When actually bringing that first AI model live in the organization, we use the Responsible AI Framework, so that we can embed it responsibly straight away. Of course, demonstrating the value of AI in the short term with a quick pilot are points easily scored. But you will have to keep emphasizing that that value is only sustainable if you do it in a responsible way."

Actual practice

The waste management industry shows high AI Readiness. On the work floor, employees regularly ask themselves questions such as: can we further optimize processes (such as crane movements, mixing processes or pouring processes)? Can we automatically recognize specific objects such as laughing gas cylinders? Can we find out who has disposed of waste incorrectly (examples include laughing gas cartridges, electronic waste or asbestos that ends up in a container) so that we can fine that person?
Waste processors do a lot of visual inspections and are already supported by cameras in that.

Conclusion has developed a Smart Waste Detection Model that can automatically detect (laughing) gas cartridges and cylinders in these camera images. This model can be further trained with images of, for example, blockages, so that processes can be optimized. The model is also suitable for Smart Document Processing: the fully automatic handling of large document flows such as invoices.