The bearing distributor: inventory up, fill rate down

Because the stock is in the wrong places, and the policy cannot tell the difference.

The rule is blind to variability and lead time. 'One month of average demand' sets safety stock proportional to the mean. It over-stocks predictable, high-volume items and under-stocks erratic ones - which is exactly backwards. Safety stock exists to cover the deviation, not the average.

The lead-time shock was never absorbed. Lead time rose 67% (45→75 days) while the policy was unchanged. For the same demand variability, required safety stock scales with the square root of lead time: √(75/45) = 1.29. The company needs roughly 29% more safety stock on volatile items just to hold the old fill rate. It is holding the old amount, at a longer exposure.

The slow movers absorb the cash. 1,900 SKUs (45% of the file) each holding a month of average demand is dead stock: under 12 units a year means a 'month of demand' is one unit - a shelf-warmer bought on a rule, not on a signal. That is where the 22% increase sits.

The top 40 SKUs stock out. These are the lumpy, high-consequence items, and the rule gives them one month of average - no buffer for a large order or a 75-day replenishment.

Result: cash tied up in the tail, service failures in the head. Total inventory and fill rate move in opposite directions, which is the signature of an undifferentiated policy.

Formula. Safety stock = z × σ over the replenishment period.

• Demand-only variability: SS = z × σ_d × √L
• With lead-time variability: SS = z × √(L·σ_d² + d²·σ_LT²)

For a 95% cycle service level, z = 1.65; 98% gives 2.05.

Use forecast error, not historical demand. With a 41% MAPE, the driver of stockouts is the gap between forecast and actual, so σ should be the standard deviation of the forecast error, which is what actually consumes buffer. That change alone usually reallocates more inventory than any service-level tweak.

Segment before computing anything. Split on value (ABC) and variability (XYZ).

Segment	Policy
A / X–Y (top 40 SKUs, 55% of revenue)	98% service level, weekly review, forecasted statistically not by sales
A / Z	Higher z, event-based buffer for known large orders
B	95%, monthly review
C / Z (the 1,900 slow movers)	80–85% service level, non-stock or drop-ship from the manufacturer, no safety stock

Explicit lead-time adjustment. Recompute every SS with L = 75 days, not 45. Expect the A-item requirement to rise by roughly 29% on the √L term while the C-item holding falls by an order of magnitude. The net inventory number should fall even as the A-item buffer rises.

The cost of service: the incremental annual inventory carrying cost required to move from today's fill rate to 95%, set against the gross margin currently lost to stockouts in the top 40 SKUs.

Build it as one table, not a slogan:

Service level	Safety stock € (segmented)	Est. lost margin from stockouts	Total cost
88% (today)	€X	€Y	€X+Y
95%	€X'	€Y'	€X'+€Y'
98%	€X''	€Y''	€X''+€Y''

The minimum of that last column is the answer, and it is almost never at 95% for every SKU. Carrying cost is roughly 20–25% of inventory value per year; the margin lost on a missed bearing order is not just that order - it is the maintenance-fit customer who sources the whole part number elsewhere.

Two supporting numbers the MD will ask for:

1. Revenue-weighted fill rate, not unit fill rate. A 95% unit fill rate that misses the 40 SKUs carrying 55% of revenue is a bad business. Report the fill rate on the A items separately; it is probably below 80%.
2. Inventory value released from the C tail. 1,900 SKUs holding one unit each, plus their min/max noise, is a concrete euro figure that can fund the A-item buffer without new money.

That framing converts a Sales vs Finance argument into an optimisation with a measurable optimum - which is the only version of this conversation that ends.