1. Which data makes a SKU worth reviewing?
Start with client ID + warehouse ID + SKU, current locations and usable status. Matching SKU text or barcodes across clients do not establish common ownership. Save the observation period, picking unit and report definitions. A practical starting sample might cover four to eight weeks, adjusted for the business cycle. Include a normal period and relevant peak conditions where possible; a quiet week alone cannot describe a promotion.
| Field group | Minimum usable record |
|---|---|
| Identity and period | Client ID + warehouse ID + SKU; source report, sample dates, current locations and usable status |
| Activity | Orders, lines, units; actual visits or explicitly labeled proxy; batch/task reference and each/case/pallet unit |
| Item and location fit | Item/package dimensions, weight, orientation and handling needs; usable space and confirmed load rating |
| Related work | Replenishment trips and person-time; search, errors, checking, waits and feedback; valid co-orders |
| Demand context | Promotions, seasonality, stockouts, new-client launch and uncertain estimates |
Keep orders, order lines, units and actual location visits separate. Sales-value ABC is not a substitute for picking-activity classification. A bulk item can produce many units in few visits, while an inexpensive accessory creates repeated small picks. If actual visits are unavailable, label order lines as a proxy and validate with a small route observation. Do not report proxy counts as measured trips.
Define a location visit before collecting it: for this pilot, it can mean one physical stop at a client–SKU location to complete the associated pick work. Several unit scans at that stop are not automatically several visits. Stops at two locations should remain distinguishable. Keep the batch or task reference and observation method with the result. If only scanner events are available, explain what they can and cannot establish about physical movement.
Batching changes that relationship. One visit can supply several orders, and one line can require more than one location. Keep the batching method with the sample so the interpretation survives a shift change. The batch picking guide covers collection and order separation; this guide uses that context to judge position candidates without redesigning the whole picking method.
Treat historical shortages and new-client launches as context. A stockout can suppress fulfilled units and make a demanded SKU look slow. Separate sales lost or work delayed by unavailable stock from the activity that actually occurred. Compare launch or promotion demand with a relevant scenario, rather than assuming the last completed orders are a complete forecast. Mark uncertain estimates and give them an early review point.
For associations, use distinct valid customer orders within a stated client and time scope. Deduplicate exports and reconnect split fulfillment records to the intended order definition. Decide how kit parents and picked components will be represented, so an assembly does not become artificial evidence of two independently chosen products. Popular items naturally co-occur often; their raw shared count alone does not prove that adjacent slots reduce work.
Check whether a customer-order association also creates shared picking work. A split order picked on different days may count once for customer demand analysis while providing no same-route opportunity. Keep both definitions rather than silently changing the denominator. Compare the proposed pair with another practical arrangement under the same order scope. A planner should be able to explain which actual work would be removed, not merely point to a large co-occurrence count.
Collect item and package dimensions, weight, orientation limits, picking unit, confirmed location space and load rating. Add replenishment trips and person-time, searching, errors, waits and worker feedback. The inventory lead verifies questionable stock or identity through the cycle counting process. Finish the candidate dataset only when the identity, unit, period and data limitations can be explained by the next reviewer.


