Lighting Retrofit at a Distribution Centre: Root Cause or Convenient Story?
A regional distribution centre picks 42,000 order lines per week and has a mis-pick rate of 0.85%, against a customer-agreed target of 0.25%. Zone A (ambient, 24 aisles) and Zone B (chilled, 9 aisles) both feed the same packing line. A Green Belt team's fishbone workshop lands on poor lighting as the leading cause, based on two pickers' comments and one walk-through at 2 p.m. The available data show day-shift mis-picks at 0.38% and night shift at 1.45%; night shift uses 58% agency staff, day shift 12%. The lighting retrofit would cost $85,000 and close Zone A for four nights. Finance wants payback inside 12 months; each mis-pick costs about $22 in return handling. A new pick-to-light system in Zone B went live six weeks ago, and the warehouse management system was updated three weeks ago.
1. Assess the lighting hypothesis. What is confounded, and what would you need to separate the explanations?
Read the worked answer
The hypothesis is plausible but currently unsupported. Shift and staffing are confounded: night shift has both the higher error rate and 58% agency staff versus 12% on days. Either poor lighting, inexperience, or both could explain the pattern - and agency staff usually work the shifts with the least experienced pickers, so the two move together. Two further confounds: the pick-to-light system in Zone B and the recent WMS update, which can change pick sequencing and screen behaviour and may have affected errors in either direction. To separate them, stratify the error data by zone, shift, picker tenure (agency versus permanent), picker, aisle, hour of day and error type (wrong item, wrong quantity, wrong location). Then compare error rates for the same tenure group across shifts: if experienced permanent night pickers err at day-shift rates, lighting or fatigue is more likely; if the gap follows tenure, training and standard work matter more. A walk-through at 2 p.m. tells you nothing about night conditions - measure illuminance in the aisles at night.
2. The team proposes retrofitting Zone A only and comparing Zone A with Zone B afterwards. What is wrong with that design, and what would you propose instead?
Read the worked answer
Zone A versus Zone B is not a valid comparison: the zones differ in product mix, temperature, aisle count, staffing patterns and now in picking technology, so any difference in error rate could be attributed to the zones themselves rather than lighting. There is also no baseline stability - both the pick-to-light rollout and the WMS update are recent, so the before and after periods are not comparable. A better design, still cheap: pilot in a matched subset of Zone A aisles, randomising which aisles get the higher-spec lighting first, with the same shift, product family and staffing mix on both arms, and enough picks per arm to see a meaningful difference (with roughly 0.85% baseline, several tens of thousands of picks per arm are needed to detect a drop to 0.25%). Fix the measurement first - an independent audit of a sample of picks with a clear definition - and hold the WMS and pick-to-light changes constant for the duration.
3. Finance asks whether an $85,000 retrofit can pay back in 12 months. What analysis do you present, and what caveat do you attach?
Read the worked answer
Work the arithmetic transparently. At 42,000 lines per week and 0.85%, the site produces about 357 mis-picks per week; at $22 each that is roughly $7,850 per week, or about $408,000 per year. Reaching the 0.25% target would remove about 252 errors per week, around $5,500 per week or $288,000 per year - so an $85,000 retrofit would pay back in well under a year if it delivered the full effect. Then attach the caveats. The $22 figure and the target both need validation; the improvement cannot be credited to lighting until the confounders are resolved; four nights of Zone A shutdown have their own cost and service risk; and if tenure or process design drives most of the gap, the retrofit will buy a fraction of the modelled benefit. Recommend funding a measured pilot first, with a decision gate: only scale the retrofit if the pilot shows a statistically credible reduction in error rate.