Ready to work, but waiting. How warehouse task assignment removed the queue for the supervisor.

A worker finishes a task, then waits.

A picker closes a task on the terminal and is ready for the next one. But the next one comes from the supervisor, assigned by hand and by eye. While the supervisor deals with another area, the worker stands still. In a warehouse with around 800 shifts a day, those pauses repeat all day long.

The supervisor is stretched too, watching queues in several areas at once and deciding who goes where. A wrong call means work piles up somewhere and a shipping deadline slips.

Step one: the next task arrives on its own.

As soon as a worker is free, a new task appears on the terminal, chosen by floor, zone and distance. The system matches people and tasks for everyone at once rather than one at a time. The first area was automated picking on the mezzanine.

Step two: people move to where the queue is growing.

The system sees where work is piling up and where a deadline is at risk. It suggests specific people based on their history: where they have worked and how productive they were. Newcomers are not moved during their first shifts, and anyone sent to an unfamiliar process goes with a mentor. The supervisor approves, changes or cancels the move.

Results, and what turned out harder than the algorithm.

In mezzanine picking, task waiting time fell by 30%. The pilot also cut manual task assignment by supervisors by 80% and improved operator efficiency by 8%. These results come from this project, not a forecast for every warehouse.

Transfer suggestions were not always accepted. The reason was human: a supervisor’s bonus depended on their own area’s numbers, so giving a strong picker to a neighbor hurt their results. A suggestion needs a clear reason, and the supervisor who gives up people should get credit for it.

This work was done at Yandex Market, and the code belongs to Yandex. It is part of our team’s experience; Hrusty is a separate product for frontline teams.