Optimised Production Planning with AI

There is a study by the WWF from 2019. It states that around 1.7 million tonnes of baked goods are disposed of in Germany every year. That is a huge amount and would probably be equivalent to the harvest of a good 400,000 hectares. Around a third of these are returns from bakeries. In other words, just over half a million tonnes. Everything that's not sold in the shops and sent back to the head office ends up in animal feed, in biogas plants or, in many cases, simply in the rubbish bin. Ultimately, baked goods are not used for their intended purpose. And let's be honest: Pig feed can certainly be produced in a more resource-efficient way than by turning the raw materials into nut bread first. That's why production planning is so important.

Traditionally, the quantities are determined in the shops. This means that employees have to plan for the next few days. And that often happens on the fly. There are two main reasons for this. Firstly, branch processes have become much more complex. Many hygiene regulations have to be adhered to, things have to be prepared and accounts have to be drawn up. And if there are five minutes free in between, a quick plan is made for the next day. Secondly, bakeries - like many other sectors - have a huge staffing problem. But planning how much bread, bread rolls and cake I need for the next day is not so easy. Planning is therefore becoming an ever greater challenge, so software simply makes sense at this point. And that's where the trend is heading

Ultimately, all the data relevant for us is produced at the cash register. When a person presses the button, a data point is created. And that's where we come in. When we start a project, we first analyse the data situation. We then customise our software until it can go further and further into automation. In this way, the system learns and, in the best-case scenario, the bakery can then take over more and more automatic data points.

Actually, the limits of our AI always lie in the data quality. Here's an example: There is a sandwich in the shop. And there's a button on the register that says "sandwich". However, the system doesn't record which of the five different grain rolls from the range is currently being used for the sandwich. The order quantities for the individual varieties can then no longer be precisely predicted. But we can see this very quickly and transparently in our user interface. So there's nothing better to be gained from these data points. In such cases, we suggest that the customer change the cash register programming. The type of grain roll is then entered when the sandwich is made. This provides better data and therefore a more accurate forecast.

I hate throwing food away. I'm also the one who eats my daughter's leftover school lunch the next morning. Don't ask me where that comes from. Maybe because as a little boy I used to spend my summer holidays on my grandma's farm. During my doctoral thesis, I worked a lot with data and wrote tens of thousands of lines of programming code to analyse it. The question of what I could do with this knowledge of data analysis, statistics and AI has occupied me ever since. And then I saw a report about food waste. There were these pictures of huge rubbish containers. There weren't one or two loaves of bread in there, but thousands. That really touched me and was ultimately the spark that ignited FoodTracks.

That's right. There was one thing I definitely underestimated: how different the processes are in the individual bakeries. From the outside, it always looks like bread rolls are being sold and that's it. But bakeries come up with a lot of ideas to be able to compete with the bake-off stations in the retail food trade and still offer the best possible quality. Some deliver the rolls already baked in the morning so that they're available immediately. Then there are the dough pieces. And there is also a so-called 'fermentation interruption', which cools dough pieces delivered the day before to a certain temperature. This ensures that they are cooked just right in the morning and can then be baked. Behind all of this is a great deal of expertise, a great deal of effort and a very high level of process complexity. This is also reflected in the data. And we had to come to terms with that first. A bread roll is not just a bread roll, but three, four, and sometimes even more items. And the quantities need to be planned correctly for all of them.

In fact, I wouldn't have thought at first that you couldn't achieve so much in this case with existing algorithms. We had to redevelop a lot of things. Like the algorithm for the sales opportunity. A classic example: I get to my bakery at 10:30 am on Sunday and the croissants are sold out. We often see this in the data. Now, of course, the question is how many croissants could have been sold before closing time? Such an algorithm was not available as a ready-made module and nobody could help us. Now it goes directly into our forecast.

We definitely face very high expectations. And ChatGPT has also contributed to this. Because it's apparent what an AI can do. It really is fantastic, almost unbelievable. It's easy to quickly think: If ChatGPT can do that, then surely I can do the same with my bakery data. And it's completely automated. But the comparison is misleading. Asking a chatbot a question is very different from fully automating production with bakery data. You see, hundreds of millions of dollars have been invested in ChatGPT to train the algorithm with clean data. Whole armies were tasked with generating training data. A bakery will never invest so much money to historically clean up its data down to the last detail. This is ultimately reflected in the forecasts. So we first have to do some educational work with bakeries in order to relativise expectations.

That's right. After all, the industry is characterised by craftsmanship. This is why we also support our customers to a certain extent with digitalisation and data-driven decision-making. I believe this is a direction that many companies want and need to take. So we're moving away from gut feeling and towards what the data says. This is important in order to remain competitive. And we've already helped a number of companies in Germany to implement this new approach. After all, just because you've done everything well on instinct for the last 30 years doesn't mean this is the best solution for the future. We're initiating major thought and change processes in the bakeries.

We already have two customers from the retail food sector. Two organic supermarket chains. This is definitely a field that we enjoy and in which we want to continue to work. First of all, we look after the bakery corners. Because we're simply professionals when it comes to bakery data. But that's a very tall order. Simply much taller than I would have expected at the beginning. That's why the knowledge and handling of data is extremely important. We're currently the market leader in the bakery sector and we definitely want to maintain and expand this position. However, it's also quite possible that we'll be looking at other parts of the product range in the future. Fruit and vegetables perhaps. Or dairy products. All those areas in which valuable resources are thrown away and for which our software is fundamentally suitable are conceivable.

Dr. Tobias Pfaff, FoodTracks,

t.pfaff@foodtracks.de

https://www.foodtracks.de/