Six months on site, scaling AI picking for e-grocery

Rohlik Group runs online grocers across Europe, including Knuspr in Germany and Gurkerl in Austria. It is rolling out Sereact’s AI picking robots for chilled and ambient goods, connected to its own logistics platform. The rollout is at 51 stations and heading for 100.

A picking robot only counts if the warehouse runs faster with it. Throughput had to go up inside a fixed, tight layout, while warehouse operations, project managers and executives each needed different answers every day.

• Moved on site for six months and took over project management when the project manager left, with three to four stakeholder meetings a day.
• Found where the pick cycle lost time and fixed it.
• Pitched automatic recovery when a robot drops an item, so nobody has to step in. It now runs on every Rohlik station.
• Wrote the Operator’s Guide, an operations and maintenance playbook that covers 98% of daily situations. It is still the first reference in Vienna and Berlin.

• Warehouse throughput up about 60%, despite the layout.
• In the chilled cell, which I took from business case into site acceptance, scan rate reached 96% and uptime stayed above 97% running 24/7 through acceptance testing.
• At gurkerl.at, a Sereact colleague reported 540 picks per hour, a 4-minute average reaction time and a 21-minute average remote fix, and called me “the final end boss” for site problems.

• Each audience needs its own answer: operations for warehouse managers, milestones for project managers, business value for leadership.
• Write the playbook on site, where the problems happen. That is why people still use it.