Rebaselining Picking Errors in a Distribution Center
The regional distribution center ships about 9,200 order lines a week across three shifts, with four pickers per shift. Packing reports a 1.8% picking error rate, up from 0.9% six months ago, and customer complaints have risen from 22 to 41 a month. The error log is kept by packing staff and records only defects caught before shipment; it has free-text entries such as 'wrong item', 'short', 'damaged' and 'missing', with no picker ID or shift. Pick lists are paper, and pickers batch 12 orders at a time across four aisles. A quick audit of 100 picks by two auditors produced 11 disagreements, mostly between 'wrong quantity' and 'damaged'. There is no count of lines picked per person, and no one has mapped the picking process. The operations manager wants a one-page dashboard by Friday and suggests blaming the newest pickers. A Yellow Belt is asked to support the Black Belt with measurement.
1. Write an operational definition of a picking defect and outline a data collection plan that would produce a trustworthy baseline. Include the unit, opportunity and key fields.
Read the worked answer
Operational definition: a picking defect is any order line where the item, quantity, condition or labeling delivered to packing differs from the pick list and order requirements, detected at pack audit, customer receipt or cycle count. Exclude errors caused after handoff (packing, loading, carrier damage) unless they trace to picking. Unit = order line; opportunity = one order line picked.
Data plan: for four weeks, audit every line at packing plus a daily random sample of 30 lines checked independently by a second auditor; also log customer-reported errors with order and picker ID. Fields: date, time, shift, picker ID, aisle, SKU, batch size, error type (wrong item, wrong quantity, damaged, missing, wrong label), detector (pack, customer, cycle count), order number, and whether the line was a look-alike SKU. Create a standard error-type pick list to remove free text; calibrate auditors on 20 lines and calculate agreement. Record denominator by picker and shift. Keep customer-reported and internal errors separate until you can reconcile duplicates; avoid double counting. Review data weekly for missing fields and outliers. This gives a defensible baseline and allows stratification by shift, picker, aisle and batch size.
2. Is the reported 1.8% error rate trustworthy? Explain what you would tell the operations manager who wants a dashboard by Friday.
Read the worked answer
No. The numerator is incomplete and the denominator is unclear. The log captures only defects caught at packing, so customer-reported errors and cycle-count finds may be missing; free-text categories mix severity and type; there is no picker or shift ID, so you cannot stratify; and no line count per person means rates by picker cannot be calculated. The 11% disagreement in the 100-pick audit is high for a simple check and shows the definitions are not reliable, so even the internal count is uncertain. The 1.8% may be a useful signal, but it is not a baseline for judging improvement or blaming the newest pickers.
Tell the manager: Friday can have a one-page 'data quality and first look' dashboard using the existing log, clearly labeled provisional, with an error-type Pareto and complaints by week. Do not publish picker rankings. In parallel, run the four-week collection and auditor calibration. Re-baseline after the definitions are stable. This protects the team from a false start and gives the Black Belt a number that can support decisions.
3. Using the process information, propose a stratification plan and explain how the process map changes what you measure.
Read the worked answer
The map shows a batch of 12 paper orders, four aisles and a staging lane, so the process creates look-alike SKU exposure and congestion that a simple overall error rate hides. Stratify by shift, picker, aisle, batch size, time of day, SKU similarity and error type. Compare error rates per 1,000 lines picked, not raw counts, because exposure differs by shift and picker. Add a short observation sample of picker travel and staging wait to see whether errors cluster where batches are split or staged.
The map also changes measurement: instead of only counting final defects, measure at the handoff points - pick-to-stage accuracy and stage-to-pack accuracy - to localize where errors originate. If most errors are found at pack but originate in the aisle, pack audit alone will mislead. Track 'found at pack' versus 'found by customer' separately.
Do not conclude that batch picking is the cause yet; the stratification may show errors concentrate in one aisle or among look-alike items. Use the data to choose a focused pilot, such as reducing batch size or separating similar SKUs, and measure lines per 1,000.