Is Inventory Turnover and Service Level a Binary Choice? —— Selecting Points on a Curve and the Three Levers to Move It

The supply chain director of an industrial consumables company was given a target by the group in 2025: to increase inventory turnover from 4.2 times to 6 times. The approach was straightforward—cut safety stock by 30% across the board. Three months later, the turnover rate reached 5.6 times, and the quarterly bonus was secured. However, the order fulfillment rate dropped from 96% to 88% during the same period, and two customers with annual procurement amounts exceeding ten million yuan switched suppliers. The company's annual sales decreased by 7%. In the following year, the group included service level in the primary performance evaluation, and the safety stock had to be restored, bringing the turnover rate back to 4.5 times. After a year of折腾 (tortuous efforts), neither number moved, and two customers were lost.

This kind of story happens in every manufacturing company, not because of poor execution, but because a "point selection problem" was treated as a "trade-off problem."

Conclusion first: Inventory turnover and service level are not two ends of a balance, but two coordinates on the same curve. Given your demand fluctuations, lead times, and procurement batch sizes, the combination of these two indicators is locked on a curve, and you can only choose points on this curve—indeed, it is a binary choice. However, the position of the curve itself can be moved: shortening lead times, reducing fluctuations, and breaking down batch sizes can achieve the same service level with less inventory. Managers need to determine not which one to protect this time, but whether they are moving points or moving the curve.

1. Why It Seems Inevitably a Binary Choice: Three Knobs in the Safety Stock Formula

The classic formula for safety stock is SS = z × σ × √L: z is the percentile corresponding to the service level, σ is the demand fluctuation during the lead time, and L is the lead time. Of the three knobs, only one is directly linked to "service level."

Criterion: Increasing the service level from 95% to 99% requires a 42% increase in safety stock, but the stockout probability only decreases from 5% to 1% (a reduction of 80%). This is the law of diminishing marginal returns—it is not a result of poor management, but of mathematics. Therefore, the last few percentage points of service level are the most expensive inventory in the company, and their benefits must be calculated based on specific instances of stockouts, not on a general sense of security.

What is truly overlooked is L. Safety stock is proportional to the square root of the lead time: reducing the lead time from 60 days to 30 days decreases safety stock by 29%; reducing it to 15 days decreases it by 50%. The same service level can be achieved without spending an extra cent, just by reducing the time it takes to complete the process. This is the segment with the highest return, but the least action, as it does not fall under the supply chain department—approval processes, quality release, transportation methods, and supplier production priorities are all managed by other departments.

2. Two Indicators Share the Same Denominator, So They Cannot Be Evaluated Separately

Turnover rate = annual cost of goods sold ÷ average inventory; service level = orders delivered as promised ÷ total orders. The numerators are entirely different, but the denominators are the same. Any action to reduce the denominator will simultaneously change both indicators, and in opposite directions. This is why "each department bears one indicator" inevitably leads to a tug-of-war.

First, the口径 (caliber) must be defined. Whether the average inventory is taken as a snapshot at the end of the month or a daily average (end-of-month inventory reduction can falsely increase turnover by 0.3 to 0.8 times), whether it includes in-transit and work-in-progress inventory, and whether it is valued at cost or selling price—these three calibers can differ by more than one turnover for the same batch of goods. Before comparing turnover rates between two business units, ensure that the denominators are the same.

More hidden is the definition of service level itself. Cycle service level (probability of no stockout) and order fill rate (proportion of demand met immediately) are not the same and can differ significantly. What customers perceive is "whether the entire order can be shipped in one go," not "whether each SKU is in stock." An average order with 5 SKUs, each with a 98% in-stock rate, has a complete order fulfillment rate of only 90.4% (0.98⁵); each SKU at 99% results in a complete order fulfillment rate of 95.1%. If a commitment is made to deliver the entire order on time based on a 99% in-stock rate for each SKU, it is mathematically impossible to fulfill.

Counterexample: A company evaluated the warehouse's "99.2% in-stock rate" based on the SKU caliber, which looked impressive, but customer complaints persisted—complaints were all about complete order fulfillment. After switching to a complete order fulfillment rate evaluation, the actual level was only 86%, and the problem was finally recognized. Choosing the wrong evaluation caliber is equivalent to hiding the problem and paying bonuses for it.

3. Pushing the Curve Inward: The Benefits and Costs of Three Levers

Expected Range: Reducing lead time (L by 30%50%) alone can decrease safety stock by 15%30% at the same service level; further converging supplier delivery windows (σ by 20%30%) can reduce it by another 10%20%; combining both typically increases turnover by 0.6～1.2 times and service level by 1～2 percentage points. Conversely, only adjusting the z value will inevitably result in opposite movements in both indicators—this is not a matter of capability, but of the formula.

This range also serves to evaluate the credibility of supplier and consulting proposals. Any proposal claiming "30% reduction in inventory, 50% reduction in stockouts" without changing σ and L is likely doing one of two things—pushing inventory upstream (VMI, consignment) or altering the statistical caliber.

4. Layered Service Level Setting: A Directly Usable Table

A uniform 99% is the most expensive and least effective approach. C-class long-tail items often account for 60% of the number of SKUs and 20% of inventory value. Paying 99% for them means diverting funds from A-class items.

Passing Line (all three must be met, failing any one is a failure):

1. A-class complete order fulfillment rate ≥ 95%, C-class can have single SKU stockouts, but they must not affect the completeness of A-class orders.
2. Inventory turnover increases by ≥ 0.5 times year-over-year, and the order fulfillment rate decreases by no more than 0.5 percentage points—exceeding this is a failure, even if the turnover target is met.
3. Obsolete inventory (no movement for 180 days) ≤ 5% of inventory value.

Three Counterexamples:

• Uniform 99% across all levels. It appears fair but actually spreads the budget evenly across long-tail items. The result is that A-class items lack the margin for dual sourcing, and a single supplier incident can cause a line stoppage.
• Making inventory turnover a personal KPI for procurement. Procurement will inevitably reduce inventory and delay orders, making the turnover look good but increasing material shortages and expedited shipping costs, leading to higher total costs. Turnover is a result metric and should only be evaluated at the organizational level, not the individual position.
• Misunderstanding service level as "zero stockouts." The true goal is "delivery as promised." Allowing C-class material stockouts but committing to delivery within three days is often acceptable to many customers, but the cost difference is significant. Delivery commitments can be designed, but inventory cannot.

5. Four Steps to Implementation

1. First, define the caliber. Use the daily average for average inventory, include in-transit inventory, and value it at cost; report both the complete order fulfillment rate and the order line fill rate for service level. Without a defined caliber, all subsequent comparisons are self-deception.
2. Layer by value and fluctuation. Use the cost of goods sold percentage and the demand coefficient of variation (CV) to create four quadrants, not by the number of SKUs—dividing by the number of SKUs will inevitably treat the long-tail as the main body.
3. Calculate the current position of the curve. Pull out the σ, L, batch size, and actual inventory for A-class items, and calculate the difference between the theoretical safety stock and the actual inventory. If the difference is over 30%, first check if the parameters are incorrect (many companies' lead time parameters are still from five years ago), rather than immediately cutting inventory.
4. Determine the lever sequence: first L, then σ, and finally z. z is the last resort because it directly buys stockout risk; after improving L and σ, the same z value can support a higher service level.

The issue of inventory turnover, on the surface, is about inventory parameters, but in reality, it is about lead times and fluctuations. Lead times and fluctuations fall under process and supplier management, respectively. Therefore, this is never something the supply chain department alone can accomplish. The supply chain's "inaccuracy" is often due to incomplete delivery data and event records from upstream, which is detailed in What is the Difference Between "Visibility" and "Accuracy" in Supply Chain? (16.2.3). The longest waiting times in the delivery cycle often occur at departmental handoffs, as discussed in Seven Breakpoints from Order to Delivery (16.2.1). The reasons why schedules cannot be adjusted and forecasts are inaccurate are covered in Why Do We Still Rely on Excel After Changing Three Production Scheduling Systems? (16.2.2) and Are Forecast Accuracy and Schedule Stability the Same Thing? (16.2.2). If you plan to use algorithms to reduce σ, first pass the six criteria in Which Categories Are Worth AI Demand Forecasting? (16.3.1).

In a Nutshell

Inventory turnover and service level are not a binary choice, but on a fixed curve, you can only choose one. The management effort lies in moving the curve inward—shortening lead times, reducing fluctuations, and breaking down batch sizes. None of these actions fall within the authority of the supply chain department alone, so it is fundamentally a cross-departmental issue, not an inventory parameter problem.

This article belongs to the "Digitalization and AI Zone · Operational Digitalization" direction, targeting operations, supply chain, and digitalization leaders, and does not involve quality system clauses.

Knowledge code: 16.2.3 Version: v20261104 Author: QTank