When the AI ​​tells the baker how many rolls to bake

Up to 20 percent of the baked goods remain on display in the evening. Some are still sold in Ruch’s “Good things from yesterday” shops or are sent to food recycling. Other ends up in the garbage. Gresens wants to use the AI ​​to reduce food waste. The forecasts should also help ensure that goods that are in particularly high demand are not sold out too early. When his team calculates how many loaves of bread, rolls or cakes will go over the counter in the branches at midday, the AI ​​sends a recommendation to their order tablet.

Will it rain tomorrow? Are Schrippen on sale? From the weather forecast to the discount campaign – the AI ​​takes all of this data into account and its forecast is pretty accurate, as Gresens reports. However, where there is more sales data from previous years, the self-learning algorithm does it easier than with cakes, for example.

It has been almost four years since the federal government launched the “Artificial Intelligence Strategy”. Train or attract AI specialists, provide a more modern computing infrastructure and accessible data pools, dovetail research and operational practice more closely – especially in medium-sized companies. That was what the grand coalition planned in 2018. Two years later, it increased AI funding by two to a total of five billion euros in order to reduce “transfer and deployment barriers”. Only just under six percent of the companies surveyed use AI, 22 percent could imagine it.

Many medium-sized companies have recognized the opportunities of artificial intelligence for logistics, production, purchasing or procurement – because customer questions can be answered automatically or sales can be forecast more precisely. In Germany, there is no knowledge problem with AI, but rather a massive implementation problem, Bitkom President Achim Berg complained at the time.

Why does AI lead such a shadowy existence in medium-sized companies? There are many reasons for procrastination. There is a lack of skilled workers, computing capacities and investments. Many approaches fail because business processes are not yet digitally recorded and the database is therefore not sufficient. And where there is no digitization, there is also no artificial intelligence. “AI is not popular with medium-sized companies,” says Kristian Kersting, Professor of Artificial Intelligence and Machine Learning at the TU Darmstadt. The topic is too abstract and fearful. “AI is not a box that can be installed somewhere as a motor, as the medium-sized engineer might be used to.”

culture of skepticism

Instead of looking at opportunities, discuss job losses or possible discriminatory data distortions. He does not experience too little skepticism and security concerns himself. “In projects, non-disclosure agreements often have to be signed before it is clear whether the problem at hand can be solved by an AI.” Sometimes an inadequate data strategy also encounters an incorrect understanding of AI. “In medium-sized companies, the thinking prevails that AI solves all problems automatically,” says Kersting. In addition, many medium-sized companies are doing well, the order books are full, so strategy development tends to take a back seat.

Tobias Greff is one of those who wants to make AI palatable to medium-sized companies. Since 2019 he has been an AI trainer in Saarbrücken at the Mittelstand 4.0 Competence Center. The Federal Ministry of Economics is promoting exchange at 20 of 27 of these centers nationwide as part of the AI ​​strategy. Greff explains possible uses and provides small and medium-sized companies with AI implementation partners. Medium-sized companies often use cloud-based solutions, for example from the start-up scene, which makes integration easier.

Of the 1,700 contacts with small and medium-sized companies that Greff counts in Saarbrücken every year, almost 30 percent are interested in the topic of AI. Machine builders want to reduce downtimes and find out early on which component is at risk of failing. “Sometimes it’s also about smaller things: for example, who can help to automatically translate the company page,” says the business IT specialist.

The more specific the application problem, the better. When a medium-sized company wanted to process its 15,000 incoming invoices a year faster, Greff referred them to a start-up that had developed an AI-based text recognition solution. Since then, the invoices have been booked automatically. At the same time, however, Greff grounded inflated expectations. “A chatbot will currently not be able to replace a customer advisor.”

Sharing experiences helps

Dirk Hecker, deputy head of the Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS, advises before starting to exchange ideas with companies that have already gained AI experience. You shouldn’t start with the most methodologically complex project and carefully consider whether it’s worth developing the AI ​​yourself or buying it.

Julia König programs AI on behalf of medium-sized companies. In 2019, the mathematician founded the start-up Ehrenmüller in Kempten. She reports that 15 employees work for her smallest customer and around 5,000 for the largest. And almost everyone sits like they do in the Allgäu. Being close – that is important to many medium-sized companies. One of her clients is a software developer for publishers who want to forecast their sales figures. Price, author, topic, month of publication, language, publisher – all of this is data that flows into the AI ​​and is intended to help publishers not to print too many books, but also not too few.

About 50 kilometers away is another of König’s customers, the food manufacturer Hochland Deutschland with brands such as Grünländer, Simply V and Patros. “We’re just learning where and how AI works for us,” says Albert Heim, who, as the head of the digital transformation department at the cheese manufacturer, is currently managing eight AI pilot projects.

An example: If the pumps block production, part of the cheese has to be thrown away. Machine and raw material data should help to identify the patterns of such failures earlier. But as soon as a part is analog in a production machine, data is missing. “It doesn’t quite work with the AI ​​yet,” says Heim. “With the predictions of how many processed cheese shells we will sell, yes.”

The knowledge gained also helps the family business to order the packaging material for the processed cheese more appropriately. With the planned expansion of the AI ​​​​to other product groups, Heim encounters another problem: data. There are plenty of them, but accessing the databases is sometimes difficult. That takes time.

“It doesn’t work without people”

At the same time, the regulatory framework is still open. The EU authorities in Brussels are working on comprehensive legal regulation. “That will affect us less. We primarily use machine and rarely personal data,” says König. But even if the AI ​​system is transparent and coordinated with the works council – do experienced branch managers like to follow a software recommendation?

Most are open to it, explains König. But what can the most accurate daily AI forecast achieve if purchasing has agreed fixed purchase figures with suppliers for the next three months? It’s about continuously developing the cooperation, says Heim.

At the Ruch bakery, they still encounter everyday reasons for deviations. If the pavement in front of a branch is torn up with a jackhammer, customers stay away. No AI can recognize that.

“It doesn’t work entirely without people,” says Managing Director Gresens. The AI ​​solution therefore does not automatically order, its people check and correct. In the long term, the software should increase sales and reduce food waste – at least a little. “It’s unrealistic to get below ten percent,” explains Gresens. After all, a branch should never look completely sold out, even in the evening. But even that they have already taught their AI.

More: What medium-sized companies have to pay for robots