Simultaneous overstock and stockout in the same catalog is not a volume signal. It is a diagnosis.
When a business is carrying too much inventory and still missing service targets, the instinct is to treat these as two problems. They are usually one problem, and adding inventory will not fix it.
The signature is specific: excess concentrated in items and locations where demand is stable, shortage concentrated where it is volatile. That pattern says the total is roughly right and the distribution is wrong.
How buffers get duplicated
Most planning systems calculate safety stock independently at each echelon, each using the same downstream demand variability. The plant holds a buffer against that variability, the regional DC holds a buffer against the same variability, and the local site holds one too. The network pays three times for one uncertainty.
Multi-echelon methods handle this by deciding where the buffer belongs rather than assuming it belongs everywhere. For most item families the answer is one or two echelons, not all of them, and the choice depends on lead times between stages and on where demand aggregates.
The lead time data problem
Before any of this works, the lead time inputs have to be real. In nearly every catalog we examine, master data lead times are quoted supplier commitments rather than measured receipts, and the measured distribution has both a longer mean and a much longer tail.
This matters more than the choice of method. Safety stock scales with lead time variability, so an understated tail produces confidently wrong targets across the entire catalog. Mining actual received dates from purchase order history is a few days of work and frequently changes the answer more than the optimization does.
- Diagnose the pattern before adding or cutting total inventory
- Refit lead times from actual receipts, not from master data
- Decide buffer placement by echelon rather than buffering every level
- Set service targets from stockout economics per segment
Keeping it from unwinding
Inventory improvements decay. Demand patterns shift, suppliers change, and parameters set once become stale within a few quarters. The programs that hold are the ones with a scheduled refresh and a named owner, not the ones with the largest initial number.
It is better to deliver a smaller improvement that persists than a larger one that quietly reverses before the next fiscal year closes.
Working on this
If this is a live question at your company rather than an interesting read, we are happy to talk it through without a proposal attached.