The Black Belt Project Nobody Needed
Crestline Distribution's 410,000 sq ft DC ships 2,700 orders and 46,000 lines a day. Over 18 months, lines per labor hour fell from 63 to 51 while mispicks rose to 1.9 per 1,000 lines from 0.7. Management added a second shift and a $600,000 voice-picking rollout, which lifted throughput 4% but not accuracy. Slotting has not been revisited since the building opened; 18% of SKUs generate 71% of picks and are scattered across all 12 aisles. Pick-team turnover is 46% a year and new hires are trained by shadowing for one day. There is no standard work documentation. The VP wants a Six Sigma black belt project with a statistical design of experiments on pick methods.
1. Is a designed experiment the right next step? What would you do first, and why?
Read the worked answer
No. A DOE studies the effect of controlled factors on a stable process. This process is not stable and the dominant variation is visible without statistics: pickers in 12 aisles travel past the fast movers, half of them are new, and none of them do the job the same way twice. Experimenting on pick method while travel path and training are uncontrolled will produce a result that cannot be reproduced.
First 60 days, in order:
- Verify the measurement. Are mispicks recorded consistently across shifts and supervisors? A jump from 0.7 to 1.9 per 1,000 may be partly a reporting change. Never improve a number you cannot trust.
- Time and map the work into travel, search, pick, and pack. In a building with this ABC profile, travel is typically the largest recoverable component - and slotting is the lever for it, not effort.
- Re-slot to velocity, with the top 18% of SKUs concentrated in the golden zone near the shipping lanes.
- Write standard work and build a certification/training matrix with a defined time-to-proficiency, replacing one-day shadowing.
- Then, on a stable process, if residual variation remains, run a DOE on pick method.
The $600k voice system improved speed 4% and accuracy not at all, which is what happens when the constraint is slotting and training rather than the pick confirmation method.
2. How would you structure the improvement and measure it so the gains aren't gamed or reversed?
Read the worked answer
Structure: pick a contained pilot - two aisles and the top ~200 SKUs - with a kaizen event to re-slot and standardize, then a control phase before scaling. Sequence matters: slotting, then standard work and training, then technology.
Measurement design:
- Baseline lines per hour and mispicks together, plotted on a control chart, not as monthly averages.
- Pair every productivity measure with a quality measure. If you measure only lines per hour, you will get exactly the mispick rate you are currently seeing, because speed and accuracy trade off at the individual level.
- Add leading indicators: travel distance per line, picks per hour, new-hire time to proficiency, turnover, and audit-based accuracy from a fixed sample rather than self-reported errors.
Guard against reversal: standard work is only durable if supervisors audit it and the slotting is refreshed on a schedule (fast movers drift). Set a quarterly re-slot trigger based on pick frequency changes above a defined threshold.
On benchmarking: resist comparing Crestline's 51 lines per hour to a published 95. Lines per hour is not comparable across buildings unless normalized for units per line, cube, ship-to profile, and automation. Normalize before you use an external number as a target, or you will chase a benchmark that describes a different operation.