Preparing the published article.
Missed Pick Rate: What to Measure Before Blaming the Robot
How to break a missed pick rate into stages and audit WMS data, labelling, and staging before you blame the robot.
Split the Missed Pick Rate Into Stages
A missed pick rate is an average of several different failures, and only some of them belong to the robot. Pull one week of completed and failed pick tasks and split each task into stages: order release to task creation, task creation to vehicle assignment, assignment to arrival at the pick face, arrival to pick confirmation, and confirmation to handover. Record the time spent in each stage and who or what owns it.
Then ask a narrow question for each stage. Was the task created when the operator expected it? Was a vehicle available at that moment? Did the vehicle arrive at the correct face? Did the picker confirm the pick in the system? Did the downstream conveyor or packing station accept the item? Each answer points at a different team.
- Stage fields: task ID, stage name, start time, end time, owning system, owning team.
- Exception fields: reason code, who closed the exception, and whether the task was retried.
- Baseline: the same stage times for tasks completed with no exception, in the same shift.
- Volume: tasks released per hour, so a stage delay can be read as a queue rather than a fault.
Audit the Data the Robot Depends On
Mobile robots act on location data, so a wrong location produces a wrong behaviour that looks like a robot fault. Walk the affected area with a printed location list and check each entry against the physical rack or staging lane. Record what you find, including the dull cases: a missing label, a label the scanner reads twice, a location recorded one bay away from its physical position, or a lane holding two different SKUs.
Sample enough locations to be useful, and choose them deliberately: the locations with the most exceptions, the locations with the fewest, and any location that appears in more than one week of exception data. Note the time of day as well, because a location that fails only under direct sunlight or only after a shift change points at lighting or at operator habit rather than at the vehicle.
- Label fields: location ID, label present, label legible, barcode type, scan success on first attempt.
- Inventory fields: recorded quantity, counted quantity, unit of measure, mixed-SKU lane flag.
- Staging fields: lane occupied when it should be empty, pallet overhanging the marked footprint.
- Handover fields: item placed on the wrong outbound lane, or placed before the system expected it.
Fix the Process Before You Retune the Fleet
Only after stage times and data quality are recorded can you attribute the loss. If the largest delay sits between task creation and vehicle assignment, the fleet is short of capacity in that hour; if it sits between arrival and pick confirmation, the picker, the label, or the item presentation is the constraint. Retuning navigation settings will not move a delay that happens before a vehicle is ever assigned.
Write a corrective action list with an owner and a date for each item: relabel a rack section, move a staging lane out of the travel path, change the trigger point at which an order releases to the fleet, or add a vehicle to the peak hour. Re-measure the same stage times two weeks later using the same report, so the comparison is like for like.
- One change at a time, with the date recorded, so the effect can be seen in the next weekly report.
- The same extract and the same shift window for every re-measurement.
- A named owner per action, and a review meeting that reads the stage report rather than the exception count.
- A decision rule: if the stage time does not move after two review cycles, escalate the constraint instead of retrying the same fix.
Keep the weekly stage report running after the immediate problem is closed. It gives you a defensible record of what the fleet actually did, and it separates a robot problem from a process problem the next time the pick rate slips.
Photo credits: the images in this article are licensed reference photographs of the relevant robot class, credited with author and licence in each caption; they are not photographs of the specific equipment discussed. Licence and author details are recorded with each image.