Start with item and unit integrity
One item requires a controlled identifier, description, base unit, pack conversion, dimensions, weight, status and handling attributes. Duplicate codes and unclear conversions create differences that counting cannot permanently fix. Assign ownership for changes and validate required fields before release.
Make every movement traceable
Receipt, putaway, replenishment, picking, return, adjustment and transfer should state what moved, how much, from where, to where, when and under whose authority. Post at the physical event or through a controlled short delay. Unrecorded movement is the main enemy of location accuracy.
Count to test controls
Cycle counting is a diagnostic process. Select items by value, risk, activity and previous error; blind-count where practical; independently verify material differences; and reconstruct transactions since the last reliable balance. Track error causes and absolute value, not only the percentage of matching lines.
Segment policy rather than applying one rule
ABC describes consumption value; XYZ can describe demand variability. Criticality, shelf life, substitution, supplier risk and repair options remain separate decisions. A slow-moving critical spare and a slow-moving obsolete item require different policy despite similar demand history.
Set buffers from evidence
Safety stock depends on demand and lead-time variation, the chosen service assumption and data quality. A reorder point combines expected lead-time demand with the buffer. Test outliers, lost-sales periods, seasonality and supplier changes before accepting the result.
Convert theory into a feasible order
EOQ balances ordering and holding assumptions, but suppliers impose MOQ, pack size and price breaks. Storage, shelf life, cash and transport add constraints. Compare feasible scenarios around the theoretical quantity rather than presenting EOQ as an automatic purchase order.
Review service and cash together
Inventory turns and ageing alone can reward under-stocking; fill rate alone can reward excessive buffers. Review service, shortages, excess, ageing, open supply and forecast error by meaningful segments. The action should be at item, supplier or policy level, not only at total-value level.
Govern overrides and exceptions
Manual overrides are sometimes correct, but every override needs a reason, owner and review date. Separate temporary event decisions from permanent master-data changes. Review repeated overrides because they reveal a policy or data assumption that no longer fits.
Put the framework into practice
Choose one process boundary, observe the work, reconcile the source record and close one repeatable control gap. Use the related tools and cases below to continue.
Written and reviewed by Amit Mehta · Reviewed 16 October 2026 · Editorial standard