Maximum flexibility in the device supply chain: Arvato accelerates workflows in the warehouse with image recognition via Microsoft Azure

The challenge: Unpack, reconfigure and prepare for shipping with zero-defect tolerance

Thousands of units of a new Microsoft device are in the warehouse waiting to be shipped. Neatly shrink-wrapped and labeled, and lined up in a row. For a specific market they can include language specific print and packaging they now need to be converted into other markets such as Italy, Spain or Germany.  Microsoft needs this flexibility in the device supply chain to be able to respond quickly and flexibly to changing demand. And Arvato can provide it. The company, a subsidiary of Bertelsmann, operates Microsoft’s international distribution centers in the US, Europe, Australia, and Asia. “The warehouse employees open the boxes, replace all the relevant components and repackage the whole thing,” explains Thorsten Daniel, Director at Arvato. “The consumer ultimately doesn't notice any of this. They expect very high quality, which is what they get.” 

Until now, the so-called operators have performed quality control manually. This time-consuming task carries a lot of responsibility, since the boxes for the devices are comparatively easy to configure. Other electronic devices are stored just a couple of shelf groups away in the warehouse. The latest batch is ready for delivery to the United Kingdom, but the tablets are needed in the German market. And not individually, but rather as a bundle with mouse, pen, and keyboard. The devices therefore get completely new packaging as well as the latest software. Quality is imperative and we look to leverage technology to enhance that capability.

As a partner to well-known companies, Arvato is an expert around the world for everything relating to modern warehousing and digital supply chain optimization as well as to logistics automation. “As part of our cloud-only strategy, we have developed technologies that simplify processes in our warehouses around the world,” says Christian Horz, CIO at Arvato. And Thorsten Daniel reveals the first details: “For quality assurance, we taught artificial intelligence how to see.”

The solution: Quality inspection with Azure image recognition and artificial intelligence

A special edition of the devices is transformed into a bundle – and the eyes of the artificial intelligence check carefully that everything is correct. The packages pass through a special tunnel on the conveyor belt. Inside are five cameras that take pictures from all sides, which are then sent to the cloud. There, the AI based on Microsoft Azure determines whether the film is fitted correctly, for example, and that the correct labels are also accurately applied. The findings are then sent on to Armada, a program developed by Arvato for accelerated and automatic control of all processes in the warehouse. It uses rules and information for this that are defined in SAP – and processes them at high speed. 

“The conveyor belts in our warehouses are controlled by Armada,” said Christian Horz. The control program assesses whether the package is ready for sale. Armada sends this assessment back to the conveyor belt, which then automatically transports the package further or, if something is wrong, sorts it out for another check. “A total of eight different AI models make this possible,” explains Thorsten Daniel.

Five seconds pass from scanning the product to categorizing it and distributing it down the conveyor belt – with nearly 100 percent detection of any errors. “Even a well-rested employee can’t do that in that time,” Thorsten Daniel said. “Our solution always evaluates in the same way and is completely objective. This way we can prevent a product that hasn't been perfectly reconfigured from leaving the warehouse.” This reduces the lead time, saves process costs, lowers the error rate – and allows the operations team to turn to more value-adding activities.

Arvato has now integrated visual inspection with computer vision into its standard logistics processes via Azure Cognitive Services. “We consolidated several processes for this and intentionally designed the solution as a platform model,” explains Thorsten Daniel, who already has other application areas in mind. He is incorporating other Azure tools incrementally – to perform assessments of packages in multiple languages, for instance. “With each use case, the platform learns and grows. Next, the AI could also check the correctness of dangerous goods labels – such as for packages containing devices with lithium-ion batteries.” And he also sees great potential in returns handling: Image recognition can be used to assess damage to returned products – which would prevent packaging that could actually be re-sent from being destroyed. This would also contribute to efforts to reduce the carbon footprint.

Arvato thinks globally. The intelligent platform solution will soon make the leap from the United States to another distribution center in Europe. And one of the Arvato warehouses in Asia could benefit from the technology going forward as well. Arvato is also studying how the solution can be used in other industries. Customers from the fashion, beauty, and healthcare industries, for example, could also benefit from this innovation in the future. 

“The project team has done a great job,” Christian Horz emphasizes. He especially highlights the “very good and cooperative partnership with Microsoft.” “The team has shown that fail fast in an agile mode and perseverance aren’t necessarily a contradiction in a pioneering project like this.”