How artificial intelligence can improve the supply chain

Another conclusion is that optimisation of current supply chains is not enough to solve the big challenges of the future. For that, radical innovations are needed, as Disney and Transiticoalition Food show.

More and more data is becoming available in supply chains, but in practice it is used only to a limited extent. How can we use AI and machine learning to realise more effective and efficient supply chains based on data? That is what consultants Lennart Bootsman and Dennis Timmers of Mobiquity will discuss.

What is so special about AI and machine learning, Bootsman introduces with a simple example: calculating the area of a figure. This requires input - the length and width of the figure - and an algorithm, in this case a simple calculation formula. "With this, we would normally program a piece of software that provides the output: the surface area. Machine learning works completely differently. In this, we train the computer (or rather the neural network) by entering lots of combinations of input and output, so that the computer 'learns' which input leads to which output."

The 'computational model' created in this way can then be used in new situations. Based on new input data, the model can predict output with a high degree of accuracy without the need to know the formula. This is especially useful in complex situations with very many factors (parameters) whose correlation cannot be determined unambiguously, making formulas inadequate. In short, in situations that occur in very many supply chains. "In such circumstances, AI provides much better predictions with greater accuracy than when applying existing techniques. Based on those predictions, supply chain professionals can make better decisions," Bootsman says.

The danger is that AI models will emerge that no one yet understands; in that case, a big black box will emerge. How can we then trust those models? How will we then know if they work well, even if they are applied in a situation the model has not seen before? To answer this, Mobiquity introduces the concept of explainable AI. By using data techniques, you can show stakeholders how the model arrives at its decisions. "This is necessary to gain confidence in the model, but also to correctly assess risks. If something goes wrong, we need to be able to identify the cause to prevent recurrence," Timmers says.

He illustrates this with an example from Google. This tech company deploys machine learning for image recognition. Based on a large number of photos of canoes, the model learns to recognise canoes. "But which image elements provide that recognition? That turns out to be mainly the paddle. Actually, Google has not built a canoe detector, but a canoe paddle detector," says Timmers. "The only solution to fix this error is to show the model lots of pictures of canoes without a paddle."

Mobiquity sees great potential for application of AI in supply chain. Now, it is already possible to support planners and operators with AI: the AI model gives suggestions and the planner decides. In the longer term, AI could also take over that role. Then self-managing factories and warehouses and possibly self-managing supply chains will emerge, says Timmers. "But we must remain critical of AI models. Even in a self-directed supply chain, we need people, among other things to monitor all the models. Before that happens, we will be a decade away."

When looking at digitalisation, there are two perspectives, both explored by the SCELP Digitalisation Working Group consisting of Edwin van den Meerendonk (Disney), Marc Moerkerken, Marc Vogels (Enza Zaden), Irke-Marjen Wiersma (Samsung) and Maurice Keuch (Mustad). The first perspective concerns looking from the present to the future, or forecasting. This is about optimisation: doing the same thing but better, for example by using AI. The question is whether this will enable us to solve the world's big problems, such as sustainably feeding more than 10 billion people in 2050. This calls for backcasting: what is the end goal we want to achieve? Which digital technologies can we use to achieve it? That is the second perspective.

According to co-founder Willem Lageweg of the Transition Coalition Food, invited by SCELP, forecasting is unsuitable to solve the food problem. Indeed, optimisation is even counterproductive. Going even further with the current form of automation leads to even more economies of scale and harmful effects for society and the environment. What we need is a different perspective. "With a group of over 100 stakeholders from different sectors and disciplines, we first tried to answer the question of what future-proof agriculture looks like. That alone led to five different images of the future. In most of them, digital technology plays an important role, but not in every future image. Technology could, for example, help to dish up a personalised diet for every human being, or very specifically use drones and robots to give each individual plant the optimal treatment."

Edwin van den Meerendonk, vice president European supply chain at the Walt Disney Company and SCELP member from the very beginning, showed how such a vision of the future can also contribute to the success of an individual company and what that means for the organisation. He explained how Walt Disney came to the decision to compete with Netflix and set up its own streaming service: Disney+. "As more and more people were cancelling contracts with cable companies, we were in danger of losing contact with consumers. That contact is important because of our consumer products and theme parks."

This decision has a huge impact on the organisation. "With Disney+, we are changing from a B2B to a B2C company. We were only doing business with cable companies, cinema operators and retailers, now with millions of consumers. That affects our infrastructure, our organisation, everything. We had a traditional organisational structure with silos, but we have turned that around. An organisation divided into silos stands in the way of a large-scale transition."

According to Prof Jack van der Veen, the example of Disney shows that we need to use both perspectives: forecasting and backcasting. "You don't change Disney's current business overnight. You have to keep improving it. At the same time, sometimes radical changes are needed to realise the dreamed-of future. The trick is to give space to the new without letting it be frustrated by the limitations of the old. And to keep the old while the new has yet to prove itself."