SAP and Migros Group Prototype Smart Ice Cream Freezer

After two weeks at SAP Switzerland’s Mode-2 Garage, Midor created a prototype of a smart chest freezer for ice cream using the Internet of Things (IoT) and machine learning.

Ueli Eggenberger, head of IT at Midor, a subsidiary of Swiss retailer Migros, which manufactures ice cream and cookies, was in for a surprise: He and a team from SAP had met at SAP Switzerland’s Mode-2 Garage, a rapid prototyping space, to discuss how to create an innovative chest freezer that orders more ice cream when supplies are low.

But the owner of a Swiss outdoor swimming pool was now saying that he wasn’t really interested in that idea. For him, a freezer that could analyze product sales at all pools and make suggestions on products to purchase based on that information would be more useful, he said. This insight was very significant for the teams from Midor and SAP at their design thinking workshop as they wanted to find out whether there was a market for their smart freezer idea.

The Mode-2 Garage, where companies work with SAP to go from idea to prototype in one week
Ueli Eggenberger (second from left) and his team hone their idea in the design phase

IoT and Machine Learning: Better Control Over Freezers and Sales

Midor leases between 4,000 and 5,000 freezers to customers. But until now, it had no way of knowing how they use them or even if they use them at all.

“What’s more, every year, Midor has to visit the customers as part of an inventory check, which is really time-consuming,” explains Daniel Kölsch, senior solution architect at SAP, who was the SAP coach overseeing prototype development.

The inventory revealed that a number of freezers were standing unused in the basement or were being used to sell other manufacturers’ products. The team wanted to find a technical solution to track the location of the freezers and identify the temperature and humidity inside them, and the ice cream products they store.

Not only would Midor then know where their freezers were, they could also tell whether they were switched on, whether the doors had accidentally been left open, and when stock needed replenishing.

The build phase begins on day two, and a team of three Midor employees and three SAP colleagues get to work. From left: Ivan Casanova (UX developer, Midor), Jennifer Kamphenkel (ML consultant, SAP), Eggenberger (at the front), Oliver Vollenweider and Filip Henggeler (both on the trainee program at SAP Switzerland), and Jan Reichert (IoT consultant, SAP)

Machine Learning: 84 Percent of Images Recognized

Whenever somebody takes an ice cream out of the smart freezer, a sensor registers the movement and a camera takes five photos to identify the product. There are other sensors that record increases in temperature and humidity as this suggests that the freezer has been left open. A GPS system saves geodata in the cloud.

Working on a prototype at this stage means training the machine to recognize only three products. After the first round of training with a batch of just 100 images, the machine correctly identified products in 84 percent of cases. In the next phase of development, more data will be used to train the algorithm, which should see its accuracy improve even further. All the information recorded is aggregated and presented on a dashboard. This means that outdoor pool businesses always know which ice creams are selling well and which ones to reorder.

“This information becomes more important as more customers use the service,” explains Kölsch. The data can be used to create benchmarks to identify which products sell best.

Sensors and a camera in the freezer track which products customers choose

Integration: Using Cloud Data in the Back-End Software

Another advantage here is that the back-end software, SAP S/4HANA, also uses this data. SAP S/4HANA saves the master data, sets prices, calculates discounts, schedules stock replenishment, and forecasts sales.

“We only need to activate two integration components to be able to create such scenarios. Afterward, you have a secure connection with your back-end system in the cloud,” explains Kölsch. These components are the connectivity service in SAP Cloud Platform and the cloud connector in SAP S/4HANA. Through the cloud, managers at Midor are able to access the dashboard, view the latest sales data, adjust forecasts, suggest the best product range to customers, and develop a forecast algorithm that draws on, for example, current weather forecasts when predicting demand. From a technical perspective, SAP software enables the user to access the back end and the dashboard. Plus, if Midor captures the freezers it leases in SAP Cloud for Customer, it won’t have to travel around checking inventory.

By hooking the smart freezer up to SAP S/4HANA, it can generate prices, discounts, replenishments, and sales information itself. Left: Casanova; right: Reichert.
The show phase on the final day features a demo of the prototype.

Next Steps: Test the Freezer, Validate the Business Case

It didn’t take long for news of the prototype to spread and generate interest from within the Migros Group. Soon, a roadshow will get underway during which Midor will offer the freezer to other companies in the industry, such as Micarna and Chocolat Frey. Midor’s marketing team is also considering putting the prototype on show at a mall to generate more curiosity at its next product launch. But before that, the prototype is being tested to make the system more stable and to gain feedback on its performance.

According to Kölsch, that is exactly the purpose of the Mode-2 Garage: “We work in short cycles, during which we develop prototypes to validate the future trends identified by our customers. This enables us to quickly identify where to invest.”

An important part of the process is finding a business case that makes taking the project further worthwhile. If the cost of checking the inventory outweighs the cost of piloting and operating the new system, then it shouldn’t be long before Midor goes live with its smart freezer.

For more information, contact