Inventory Optimization and Safety Stock

Inventory is working capital converted into physical stock. The goal of inventory optimization is to hold the minimum stock necessary to achieve a target service level — not the maximum stock that prevents every possible stockout.

Inventory Types

Type	Definition	Driver
Cycle stock	Average inventory between replenishment orders	Order quantity and order frequency
Safety stock	Buffer against demand and supply variability	Service level target and lead time variance
Pipeline / in-transit	Stock moving between nodes	Lead time × average daily demand
Anticipation stock	Built ahead of seasonal peak or planned event	Forecast of surge demand
Hedge stock	Protection against supply disruption (single source, tariff risk)	Risk assessment
Obsolete / dead stock	No demand in 12+ months	Poor new product planning, lack of write-off discipline

Cycle stock and safety stock together equal average on-hand inventory. Everything else is a cost of poor planning.

Economic Order Quantity (EOQ)

The order quantity that minimizes the sum of ordering cost and holding cost:

EOQ = √(2 × D × S / H)

Where:
  D = Annual demand (units)
  S = Cost per order (setup/ordering cost, $)
  H = Annual holding cost per unit ($ — typically 20–30% of unit cost)

EOQ is a useful baseline but assumes constant demand and instantaneous replenishment. In practice, supplier MOQs, truck load constraints, and promotional cadences override the pure EOQ.

Reorder Point (ROP)

The inventory level that triggers a replenishment order:

ROP = (Average daily demand × Lead time in days) + Safety Stock

When on-hand inventory hits the ROP, an order is placed. The safety stock component is the buffer that absorbs variability during the replenishment lead time.

Safety Stock Formula

SS = Z × σ_LTD

Where:
  Z     = Service level Z-score (see table below)
  σ_LTD = Standard deviation of demand during lead time

σ_LTD = √(LT × σ_d² + d² × σ_LT²)

Where:
  LT    = Average lead time (days)
  σ_d   = Standard deviation of daily demand
  d     = Average daily demand
  σ_LT  = Standard deviation of lead time (days)

Both demand variability (σ_d) and lead time variability (σ_LT) contribute to safety stock requirements. Reducing lead time variability (supplier reliability, lead time compression) is often more effective than reducing demand variability.

Service Level Z-Scores

Target Service Level	Z-Score	Safety Stock Impact
90%	1.28	Baseline
95%	1.65	+29% vs. 90%
98%	2.05	+60% vs. 90%
99%	2.33	+82% vs. 90%
99.5%	2.58	+102% vs. 90%

Moving from 95% to 99% service level increases safety stock by ~40%. The marginal cost of each additional percentage point of service level increases non-linearly — assess the cost of a stockout vs. the carrying cost of additional safety stock before committing to >99%.

ABC/XYZ Policy Matrix

	X (stable)	Y (variable)	Z (erratic)
A (high value/volume)	Lean SS; tight min/max	Moderate SS; frequent review	High SS or make-to-order
B (medium)	Standard SS formula	Standard SS + judgment	Wide SS or min/max
C (low value)	Generous SS acceptable	Generous SS or periodic review	Min/max or zero stock

A/X items justify the most analytical rigor. C/Z items should be reviewed for deletion before optimizing safety stock.

Replenishment Review Policies

Policy	Mechanism	Best For
Continuous review (s, Q)	Order Q units when inventory hits reorder point s	High-value A items; automated WMS replenishment
Periodic review (R, S)	Review every R periods; order up to level S	Lower-value items; supplier order consolidation
Min/max (s, S)	Order up to S when inventory hits s	Simple systems; B/C items
Vendor-managed inventory (VMI)	Supplier owns the replenishment decision	Strong supplier relationship; stable demand

Inventory Turns and Benchmarks

Inventory Turns = Annual COGS / Average Inventory Value
Days of Supply (DOS) = 365 / Turns

Sector	Typical Turns	Best-in-Class
Grocery / FMCG	20–30×	35×+
General retail	4–8×	10×+
Industrial distribution	4–6×	8×+
Consumer electronics	6–10×	15×+
Apparel	3–6×	8×+

Low turns are a symptom — of poor forecasting, long lead times, high minimum order quantities, or product proliferation. Diagnosis before prescribing inventory reduction targets.

Inventory Carrying Cost

Annual carrying cost as % of inventory value:

Component	Typical Range
Capital cost (cost of money)	8–15%
Storage space	2–5%
Obsolescence / shrink	2–6%
Handling (cycle counts, moves)	1–3%
Insurance	0.5–1%
Total	15–30%

Use 20–25% as the default carrying cost rate for EOQ and make-vs-buy trade-off calculations unless a more precise figure is available.

---

Service Level Types: CSL vs. Fill Rate vs. Ready Rate (Course 4.5 Depth)

The phrase “service level” means different things depending on context. Using the wrong definition leads to systematically wrong safety stock calculations.

Type 1 — Cycle Service Level (CSL): The probability of no stockout occurring during a replenishment cycle. If CSL = 95%, then 19 out of 20 replenishment cycles will not experience a stockout. This is the service level that maps directly to the Z-score safety stock formula. CSL of 95% → Z = 1.65.

Type 2 — Fill Rate: The fraction of total demand satisfied from on-hand inventory (no backorder, no lost sale). A 95% fill rate means 95% of all units ordered are available immediately. Fill rate ≥ CSL for the same safety stock level — a 95% CSL produces a fill rate above 95% (sometimes significantly above, depending on demand pattern). Conversely, achieving 95% fill rate requires less safety stock than achieving 95% CSL.

Type 3 — Ready Rate: The percentage of time at least one unit is available on hand (inventory > 0). Used primarily for spare parts and service part applications where any availability is the standard (binary: available/unavailable). The ready rate is a weaker standard than fill rate for high-demand items — you can have a 100% ready rate and still have chronic stockouts if you always have one unit but never enough.

Practical implication: When a customer says “we need 99% service level,” you need to ask: 99% of what? CSL, fill rate, and ready rate produce very different safety stock targets. Misunderstanding costs real money.

The Safety Stock Hockey Stick

The relationship between CSL and safety stock is non-linear. The marginal cost of each additional percentage point of service level increases dramatically above 95%:

CSL	Z-score	SS relative to 90% baseline
90%	1.28	1.0×
95%	1.65	1.3× (+29%)
98%	2.05	1.6× (+60%)
99%	2.33	1.8× (+41% over 95%)
99.5%	2.58	2.0×

Moving from 95% to 99% CSL requires 41% more safety stock. Practical example: if you’re carrying $10M in safety stock at 95% CSL, reaching 99% requires adding approximately $4M in safety stock at identical demand variability and lead times — plus the 20–30%/year carrying cost on that additional $4M.

This math makes the service level policy decision an explicit financial trade-off, not a service aspiration. The right service level for a given SKU is the point where the marginal cost of additional safety stock equals the marginal benefit of the additional service (stockout avoidance).

The Normal Distribution Assumption: When It Fails

The standard safety stock formula assumes demand during lead time is normally distributed. This assumption is convenient but frequently wrong:

Demand reality: Real demand distributions are typically:

• Right-skewed: Positive demand spikes (promotions, bulk orders, hoarding) are more extreme than negative depressions. The normal distribution is symmetric; real demand isn’t.
• Heavy-tailed: Extreme demand events (3–4 sigma) are more common in real supply chains than the normal distribution predicts.
• Zero-inflated: Intermittent demand items have many zero-demand periods, violating the continuous distribution assumption entirely.

When the normal assumption breaks down: when the coefficient of variation (CV = standard deviation / mean demand) exceeds 0.5–0.75. Above this threshold, the distribution is sufficiently non-normal that using the normal-distribution Z-score will systematically underestimate required safety stock.

Alternatives:

• Empirical distribution: Use historical demand data directly without fitting a parametric distribution. Appropriate when you have sufficient history (2+ years of weekly data).
• Gamma distribution: Handles right skew and is always non-negative. Good for moderately variable demand.
• Lognormal distribution: For highly skewed demand (CV > 1.0). Heavy-tailed and right-skewed; appropriate for items with rare large orders.
• Negative binomial: For count data (discrete units, intermittent demand); naturally handles zero-inflated demand.

Most planning platforms allow distribution selection by SKU or segment. Defaulting to normal distribution for the entire portfolio is a data quality issue masquerading as a modeling choice.

Forecast Accuracy Benchmark: The Compound Accuracy Problem

World-class forecast accuracy at the product/week level is approximately 70% (Coca-Cola benchmark for a company of that complexity). This sounds low but represents best-in-class for a broad portfolio.

The compound accuracy problem: integrated system accuracy is the product of accuracy at each layer:

• 70% forecast accuracy × 85% supply plan execution × 95% inventory record accuracy = 57% integrated accuracy

A 57% integrated system accuracy means roughly 43% of individual planning events (replenishment orders, allocation decisions, deployment decisions) are based on at least partially incorrect information. This compounds into stockouts, excess inventory, and expediting cost that appears “unexplainable” to teams that only look at one layer at a time.

The implication: improving forecast accuracy from 70% to 80% while supply plan execution stays at 85% and inventory accuracy stays at 95% moves integrated accuracy from 57% to 65% — a meaningful improvement, but one that demonstrates how each layer constrains the others.

Inventory Types: Full Taxonomy

The five operationally meaningful inventory types (the sixth — obsolete/dead stock — is a planning failure, not a policy):

Type	Definition	Planning driver
Cycle stock	Average inventory between replenishment orders	EOQ and order frequency
Safety stock	Buffer against demand and lead time variability	Service level target × σ_LTD
In-transit / pipeline	Stock moving between nodes during lead time	Average daily demand × lead time
Anticipation stock	Pre-built ahead of seasonal peak or planned event	Shape of the seasonal forecast
Hedge stock	Protection against supply disruption or price volatility	Risk assessment; single-source exposure

Anticipation stock is a planned investment; hedge stock is an insurance decision. Both require explicit authorization — they don’t emerge from safety stock formulas.