SUPPLY CHAIN MANAGEMENT

Lecture 5:
Bullwhip Effect
& Risk Pooling

Prof. Seung Jun Lee

Chung-Ang University | Section 03 | English Instruction

sjlee1@[Link] | Office: 310-1231

Office Hours: Wednesday by Appointment

Learning Objectives

Explain the Bullwhip Effect (BWE) — what it is, why it occurs, and why it matters for supply chain
01 performance

02 Identify the four root causes of the BWE and evaluate real-world countermeasures

Understand the concept of Risk Pooling and how aggregating demand reduces variability and safety stock
03 requirements

04 Apply Risk Pooling formulas to calculate safety stock savings from centralization

Explain postponement (delayed differentiation) and its connection to risk pooling — using the Benetton
05 case

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

01 The Bullwhip Effect

Definition · Why It Matters · Amplification Mechanism

Supply Chain Management — Chung-Ang University

What Is the Bullwhip Effect?

Definition | The Bullwhip Effect is the phenomenon where demand variability amplifies as it moves upstream through a
supply chain — small fluctuations in consumer demand become large swings in supplier orders.

Very High
Variability

High
Variability
Moderately
Amplified
Slightly Manufacturer
Low Amplified Distributor Orders
Variability Wholesaler Orders
Retailer Orders
Consumer
Orders
Demand
← Consumer end Manufacturer / Supplier en d →

The bullwhip 'cracks' at the manufacturer — the furthest point from real consumer demand

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Why the Bullwhip Effect Matters

Distorted information flowing from retailers to manufacturers leads to three major categories of waste and cost.

Excess Inventory Poor Customer Service Distorted Production Plans

Manufacturers build up buffer stock to Paradoxically, supply chains can Factories ramp capacity based on inflated
protect against the amplified swings they simultaneously hold excess inventory in order signals, then face sudden
see in orders — most of which does not the wrong places while stocking out in cancellations when the bubble bursts. This
reflect real consumer demand. others. The mismatch between perceived creates expensive overcapacity followed
demand and actual demand creates by layoffs.
2025: During COVID (2020–21), every shortfalls.
semiconductor company was hit with panic 2025: Korean battery makers (Samsung
orders 3–5× real demand. TSMC and 2025: PlayStation 5 shortage lasted 2+ SDI, LG Energy) over-invested in EV battery
Samsung SDI both over-expanded capacity years despite Sony having supply — they capacity in 2022–23 based on order signals
in response. misread demand signals inflated by retailer later cancelled as EV demand growth
stockpiling. slowed.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Bullwhip Effect — Industry Evidence (2025)
The BWE has been documented across virtually every industry. Post-COVID supply chain disruptions provided a vivid real-
world laboratory.

Event: 2020–2022: Global chip shortage Key lesson:

Semiconductors Bullwhip triggered by rationing gaming —
Panic ordering from automotive (GM, Ford) and consumer electronics customers ordered 5× their need to guarantee
TSMC / Samsung / companies inflated apparent demand 3–5×. TSMC planned $100B+ allocation.
Intel expansion. By 2023, chip inventories ballooned and prices collapsed.

Event: 2020: Toilet paper / hand sanitizer Key lesson:

Consumer Products Classic demand forecast updating + order
Retail sell-through of toilet paper rose ~20% during early COVID batching amplification. End-consumer demand
Procter & Gamble / lockdowns. Retailer order signals sent to P&G suggested demand had barely changed.
Coupang tripled. Manufacturers scrambled. Empty shelves lasted months
despite stable long-run consumption.

Event: 2022–2024: EV demand reversal Key lesson:

EV Batteries Long lead times (18–24 months for new battery
Automaker orders for battery packs surged 2021–2022 based on EV plant) made BWE especially damaging — by
LG Energy / Samsung growth projections. Battery makers invested billions in new capacity. the time capacity arrived, orders had
SDI By late 2023, Ford/GM/VW cut EV production targets. Cancellations evaporated.
flooded in.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

02 Causes of the
Bullwhip Effect

4 Root Causes · Diagnosis · Real Examples

Supply Chain Management — Chung-Ang University

The Four Root Causes of the Bullwhip Effect

01 Demand Forecast 04 Rationing &

02 Order Batching 03 Price Fluctuations
Updating Shortage Gaming

Each stage of the supply Rather than ordering Promotions and quantity When supply is short,
chain independently continuously, companies discounts incentivize forward manufacturers allocate
forecasts demand and consolidate orders into buying — purchasing more proportionally to orders.
adjusts orders upward batches (weekly, monthly) to than currently needed when Knowing this, customers
whenever it sees higher- save on ordering costs or to the price is low. This creates over-order to guarantee
than-expected orders from fill trucks. The periodic order feast-and-famine demand supply. When the shortage
downstream. Small signal spike creates an artificial cycles for manufacturers. ends, orders collapse
changes cascade into large demand pattern upstream. suddenly — leaving the
order swings. manufacturer with no signal
of real demand.

Understanding the root cause is essential — the right countermeasure depends on which cause is driving the amplification in your supply chain.
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Cause 1 — Demand Forecast Updating
Each supply chain stage independently updates its demand forecast when it receives orders from the stage below — and
adds safety stock on top. This double amplification is the most fundamental BWE cause.

Consumer Retailer Wholesaler Manufacturer

Demand Forecast Forecast Receives

→ → →
Sees 110 orders Sees 130 orders Sees 160 orders
Actual demand:
→ forecasts 115 → forecasts 138 → plans for 180
Mean=100, σ=10
+SS = orders 130 +SS = orders 160 +SS = produces 200

Countermeasure: Share point-of-sale (POS) data directly with all supply chain partners. Use centralized demand sensing (AI-powered)
rather than independent forecasting at each stage.
Order amplification as signal moves upstream →
2025 example: Walmart shares real-time POS data with all suppliers via Retail Link. Coupang's vendor portal gives suppliers live sell-through data.
Amazon Vendor Central provides weekly demand signals.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Cause 2 — Order Batching
Companies do not order every day — they wait and accumulate demand, then place one large order. This creates an
artificial demand spike followed by silence, which the upstream supplier misreads as volatile demand.

Why companies batch orders Supplier's view of demand

• Fixed ordering cost S makes it cheaper to order infrequently and in bulk (EOQ
logic)
• Transportation costs — full truckload (FTL) is much cheaper per unit than LTL

• Supplier promotions and volume discounts reward large periodic orders

• Monthly purchasing cycles — corporate buying departments review at fixed
intervals

Wk1 Wk2 Wk3 Wk4 Wk5 Wk6 Wk7 Wk8

Supplier sees: 0,0,0,SPIKE,0,0,0,SPIKE
— looks like unstable demand!

Countermeasure: EDI and e-procurement reduce ordering costs toward zero, enabling continuous ordering. Amazon's automated replen ishment
orders daily or even hourly — eliminating batch spikes entirely.
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Cause 3 — Price Fluctuation & Forward Buying
When manufacturers offer temporary promotions or quantity discounts, customers buy far more than their current needs
— creating an artificial demand surge during promotions and a demand void afterward.

How Forward Buying Works Evidence & Countermeasures

1. A retailer normally sells 1,000 units/month of a grocery item In the US grocery industry, ~80% of distributor-to-manufacturer
transactions involve forward buying (Kurt Salmon Associates). In
2. Manufacturer offers 20% discount for orders placed this month Korea, Lotte Mart and E-Mart promotions create similar BWE
patterns with their FMCG suppliers.
3. Retailer orders 4,000 units (4-month supply) to capture the
discount

4. Manufacturer sees 4× spike in demand → ramps production EDLP — Every Day Low Pricing
Walmart pioneered this. Stable prices eliminate the incentive to
forward buy. Suppliers benefit from smooth, predictable orders.
5. Next 3 months: retailer orders 0 (drawing down inventory)
Limit purchase quantities
6. Manufacturer misreads this as demand collapse → cuts production Allocate a maximum per customer per order to prevent hoarding during
promotions.
7. Consumer demand was flat throughout — the entire cycle was
artificial Activity-based costing
Help distributors understand the true cost of forward buying (capital
tied up, storage, spoilage risk).
2025: Korean convenience store chains (GS25, CU) run weekly promotions — suppliers report order amplification of 3–5× during promo weeks. GS25
is piloting AI-based order-smoothing to counteract this.
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Cause 4 — Rationing & Shortage Gaming
When manufacturers cannot meet all demand, they often allocate proportionally to orders received. Customers quickly
learn this and inflate their orders — giving the manufacturer completely false demand signals.

1 Shortage occurs | Manufacturer can only supply 50% of total orders due to a production constraint or raw material shortage.

2 Proportional allocation | Manufacturer allocates 50% of each customer's order. A customer who orders 100 receives 50.

Customers learn & game | Next cycle, every customer doubles their order — knowing they'll only get 50%. A customer who
3 needs 100 now orders 200.

Demand signal destroyed | Manufacturer now sees 200 as 'demand' — massively overstates true need. Production ramps
4 wrongly.

Shortage ends — orders collapse | When supply normalizes, customers stop gaming. Orders drop suddenly to actual need.
5 Manufacturer is left with excess capacity.

Countermeasure: Allocate based on historical sales (not current orders) — gaming disappears because inflating orders no longer helps. Share
inventory and capacity data openly with customers so they can see the real supply situation.
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Countermeasures — Matching Solution to Cause
The right countermeasure depends on the root cause. Applying the wrong solution wastes resources. Use this framework to
diagnose and prescribe.
Cause Countermeasures

Demand Forecast • Share POS data directly with all partners • Use AI demand sensing (centralized) • Implement VMI — let supplier see and
Updating manage inventory • Reduce lead times to shorten forecast horizon

• EDI & e-procurement to eliminate ordering costs • Logistics outsourcing to enable small frequent shipments • Amazon -
Order Batching
style auto-replenishment (daily or hourly) • Mixed-SKU truckloads to maintain FTL efficiency

• Every Day Low Pricing (EDLP) — Walmart model • Limit order quantities per customer per promotion • Activity-based
Price Fluctuation
costing to reveal true cost of forward buying • Continuous replenishment contracts with price stability

Rationing & • Allocate based on past sales — not current orders • Share real-time inventory and capacity data openly • Use reservation
Shortage Gaming systems for critical components • Penalize order cancellations contractually

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

03 Risk Pooling

Demand Variation · Pooling Math · Safety Stock Reduction

Supply Chain Management — Chung-Ang University

Measuring Demand Variability
Before we can talk about pooling demand, we need two tools to measure how much demand varies. These are essential
inputs to all safety stock calculations.

Standard Deviation (σ) Coefficient of Variation (CV)

σ = √[ Σ(xᵢ − μ)² / n ] CV = σ / μ

Absolute measure of variability around the mean. Tells you how Relative measure of variability — standardized to the mean.
many units demand typically deviates from average. Allows comparison of variability across products with different
average demand levels.
Product A: mean demand = 1,000 units/week, σ = 200 units Product A: CV = 200/1,000 = 0.20 (20% variation)
Product B: mean demand = 1,000 units/week, σ = 50 units Product C: mean = 50 units/week, σ = 20 units → CV = 0.40 (40%
→ Product A is 4× more variable in absolute terms. variation)
→ Product C is relatively MORE variable despite lower absolute σ.

Application: Used to classify products: CV < 0.2 = low, 0.2–0.5 =

Application: Used directly in safety stock: SS = z × σ_L
medium, > 0.5 = high variability

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Risk Pooling — The Core Concept

Risk Pooling | Demand variability is reduced when you aggregate demand across multiple products or locations. High
demand from one customer or region is offset by low demand from another — the peaks and troughs cancel out. As total
variability falls, required safety stock falls too.

Mathematical Foundation Real-World Applications

For n independent demand streams with standard DC Consolidation

deviations σ₁, σ₂, ..., σₙ: Merging two regional warehouses into one central DC pools the
demand from both regions. Amazon has systematically
consolidated DCs to exploit this effect at scale.
Pooled σ = √(σ₁² + σ₂² + ... + σₙ²)

This is ALWAYS less than the sum (σ₁ + σ₂ + ... + σₙ) Product Standardization
Using one universal component instead of two specialized ones
pools demand across both products. Apple A-series chip used
The gap between Quick
pooled σ and the
Example: sum of σs = the safety
2 Streams across iPhone, iPad, MacBook — massive pooling benefit.
stock saving from pooling.
Stream A: σ = 30 units Stream B: σ = 40 units Flexible Manufacturing
A factory that can switch between products A and B pools their
Separate: Total σ = 30 + 40 = 70 units of safety stock demand — excess demand for A can be covered from B's capacity,
and vice versa.
Pooled: σ_pool = √(30² + 40²) = √(900+1600) = √2500 = 50 units

Saving: 70 − 50 = 20 units (29% reduction!)

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Risk Pooling — Safety Stock Formulas
Pooling reduces the standard deviation of combined demand — which directly reduces the safety stock required to achieve
a given service level.

Safety Stock Formulas with Risk Pooling

Each location carries its

Separate SS (FOQ) SSᵢ = z × √L × σᵢ for each location i own SS independently

One central location covers

Pooled SS (FOQ) SS_pool = z × √L × σ_pool where σ_pool = √(σ₁² + σ₂² + ... + σₙ²) combined demand

Review period also covered

Separate SS (FTP) SSᵢ = z × √(L+T) × σᵢ for each location i — larger SS per location

Same pooling benefit

Pooled SS (FTP) SS_pool = z × √(L+T) × σ_pool applies to FTP model

The formula σ_pool = √(Σσᵢ²) assumes demand streams are independent. If demands are positively correlated (move together), pooling
provides less benefit. Negative correlation provides more benefit than independence.
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Risk Pooling — Two Ways to Win
Risk pooling creates value in two distinct ways — you can use it to cut costs OR to improve service. Most firms use a
combination of both.

Option A: Same Service Level, Less Inventory Option B: Same Inventory, Higher Service Level

Same SL → ↓ Inventory Same Inventory → ↑ Service Level

Keep your target service level constant (e.g., 95%) after pooling. Keep total inventory investment constant after pooling.
Result: Required safety stock drops because pooled σ is smaller. Result: With the same units now covering a lower-variability
Benefit: Lower inventory investment, lower holding cost, higher demand stream, your in-stock probability rises.
ROI. Benefit: Better customer satisfaction, fewer lost sales.

Example: Two DCs each hold 100 units of safety stock (total 200). Example: Two DCs each hold 100 units of SS (total 200).
After centralization: one DC holds only 140 units — a 30% After pooling into one DC with 200 units total:
reduction. Service level rises from 95% to 98%+ because σ is now smaller.
Same 95% service level at 30% lower inventory cost.

2025: Amazon uses both strategies — pooling demand across Prime members reduces inventory, while the savings fund faster delivery promises
(raising effective service level).
Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University
Risk Pooling — Samsung Galaxy Tablet Components PROBLEM 1

Scenario | Samsung Electronics produces two Galaxy tablet models: Tab S9 uses memory chip A (mean daily demand = 100
units, σ = 30 units) and Tab S9+ uses memory chip B (mean daily demand = 300 units, σ = 40 units). Both chips currently
come from separate suppliers. Inventory is monitored continuously. Lead time = 1 day. Target service level = 98% (z = 2.05).
Assume demands are independent.

(a) Calculate the safety stock required for chip A and chip B separately. What is the total safety stock?

Samsung's component engineers propose redesigning both tablets to use the same unified memory chip, pooling
(b)
demand. Calculate the pooled standard deviation and the new total safety stock.

(c) How many units of safety stock does Samsung save by using a common chip? Express this as a percentage reduction.

What is the trade-off Samsung must consider when switching to a single common chip? Is pooling always the right
(d)
decision?

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Risk Pooling — Coupang DC Consolidation PROBLEM 2

Scenario | Coupang currently operates two regional fulfillment centers — one in Seoul (FC-A) and one in Busan (FC-B). For
a popular electronics item, daily demand in Seoul ~ N(μ=30, σ=4) and in Busan ~ N(μ=12, σ=3). Delivery lead time from
supplier is 3 days. Coupang uses continuous review (FOQ). Target service level = 95% (z = 1.65). Coupang is considering
consolidating to a single national DC.

Q1 What is the safety stock Coupang must carry in FC-A and FC-B separately? What is the total system safety stock?

If Coupang consolidates to one national DC (same lead time = 3 days, same 95% service level), what is the new
Q2
safety stock?

Q3 How many units does Coupang save? What is the % reduction?

The national DC is located in Icheon (between Seoul and Busan). Lead time to customers increases from 3 days to 4
Q4 days due to the extra distance. Recalculate safety stock for the consolidated DC at 95% SL with L = 4 days. Is
consolidation still worth it?

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

04 Service Level Strategy

Differentiated SL · Profit Optimization · Portfolio Approach

Supply Chain Management — Chung-Ang University

Service Level — Strategic Differentiation
Not every product deserves the same service level. Setting the right service level per SKU — based on its commercial
characteristics — is how leading retailers maximize profit while controlling inventory cost.

High Profit Margin High Volume Low Demand Variability Short Lead Time

→ Higher SL → Higher SL → Higher SL → Higher SL

A stockout on a high-margin High-volume items are more Predictable demand means the Short lead time means safety
item loses more profit per unit visible. A stockout affects more cost of additional safety stock is stock is small even at high z-
than on a low-margin item. customers and creates more lost low (small σ). You can afford a values. You can achieve 99% SL
Worth carrying extra inventory. sales revenue in absolute terms. high SL cheaply when demand is without much inventory.
stable.

Apple AirPods Pro: 60%+ margin GS25 convenience stores: Household staples (rice, cooking Domestic supplier (2-day lead):
→ target 99% SL ramyeon (top 3 SKU by volume) oil): low CV → high SL cheap 99% SL feasible at low cost
Coupang Rocket: house-brand → 99% SL Seasonsal or trend items: high Imported goods (30-day lead):
basics: 20% margin → target Specialty energy drink: low CV → high SL very expensive same 99% SL requires 4× more
90% SL volume → 85% SL acceptable SS

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Service Level — Profit Optimization Framework
Rather than setting a single target service level for all products, leading firms optimize service level per SKU to maximize
expected profit. The result: high-margin stable items get >99% SL, while low-margin volatile items get <90%.
Premium Zone Lean Zone
SL > 99% SL < 90%
High margin + High volume + Low variability + Short lead time Low margin + Low volume + High variability + Long lead time

Examples: iPhone flagship, Samsung OLED TV, LG Gram laptop Examples: Seasonal specials, niche electronics, high-fashion items

Standard Zone
Consider Dropping
SL 90–99%
Medium characteristics across most dimensions Negative expected profit at any realistic service level. These items
may not belong in the assortment at all.
Examples: Mid-range appliances, standard clothing SKUs, most
FMCG Action: Renegotiate margins, reduce assortment, or use VMI.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

05 Postponement &
Modular Design

Delayed Differentiation · Benetton · 2025 Examples

Supply Chain Management — Chung-Ang University

Postponement — Delayed Differentiation

Postponement | Delay the point at which a product becomes differentiated (customized for a specific market, customer, or
configuration) for as long as possible — keeping it in a generic, poolable form until demand becomes clearer. This is risk
pooling applied to product design and process sequencing.

Form Postponement Time Postponement Process Postponement

(Product modularity) (Delayed delivery) (Operations reversal)

Design products with common platforms Hold inventory centrally and only ship to Resequence production steps to move the
and add differentiation late. specific locations when demand is 'differentiating' step later in the process,
confirmed. after demand uncertainty is reduced.
Example: HP LaserJet printers shipped to
Europe in a generic 'world product' form — Example: Zara keeps most inventory at its Classic example: Benetton reversed dyeing
power adapters, language packs, manuals Arteixo, Spain mega-warehouse and ships and knitting — knit first in undyed yarn,
added at the regional DC based on actual twice per week to stores worldwide — then dye to the colors that are actually
orders. Inventory of one generic SKU responding to real sell-through data rather selling. (Full case on next slides.)
instead of 20 market-specific SKUs. than speculative forecasts. Competitive
advantage: 2-week lead time from design
to store.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Case: Benetton — The Original Postponement Pioneer
Benetton is a world leader in knitwear with thousands of stores globally. Their process reversal — changing the sequence of
knitting and dyeing — is one of the most famous examples of postponement in supply chain history.
The Problem: Fashion color uncertainty

Traditional process: Dye yarn first → Knit garments in specific colors → Ship to stores. Problem: Colors are chosen months before the
season. If teal is hot and red disappoints, Benetton is stuck with thousands of red garments and cannot respond.

BEFORE: Traditional Sequence AFTER: Postponement (Benetton)

1. Dye yarn (commit to color early) 1. Knit garments in undyed (greige) yarn — generic

2. Knit garments in specific colors 2. Wait to observe early-season color trends

3. Ship to regional DC 3. Dye garments to match what is ACTUALLY selling

4. Distribute to stores 4. Ship dyed garments to regional DC

5. Markdowns on wrong colors 5. Fewer markdowns — right colors in right quantities

Key enabling insight: Knitting lead time (weeks) >> Dyeing lead time (days). So knit early (long step) in undyed form, dye late (short step) when
demand is known. Risk pooling benefit: one grey garment inventory serves all colors.
2025 equivalent: Nike iD — shoes manufactured in generic form at factories, customized with colors, logos, and text at regional hubs after customer
order is placed.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Postponement — 2025 Applications
The postponement principle — delay differentiation until demand is known — appears across very different industries and
forms in 2025.
Apple — iPhone — Component Commonality Toyota — Flexible Manufacturing — Process Postponement

The Apple A18 chip is used across iPhone 16, iPad Pro, and MacBook Air. Toyota's flexible assembly lines can produce multiple models (Camry, RAV4,
One chip SKU serves three product lines — massive demand pooling. Apple Highlander) on the same line without retooling. The line is
manufactures A18 chips in large uniform batches at TSMC, then 'undifferentiated' by model until the customer order determines which
differentiates by product line at final assembly. body goes down the line.

Result: Lower per-unit chip cost, higher negotiating power with TSMC, Result: Aggregate demand across models reduces effective variability — a
fewer supply shortages. demand drop for one model is partially offset by others.

Hyundai Paint Shop — Form Postponement Zara — Time Postponement

Hyundai produces car bodies in unpainted form (grey metal) and moves Zara holds ~50% of inventory in semi-finished or unallocated form at its
them to the paint shop based on dealer orders. Color is added late in the central Spanish warehouse, shipping twice per week based on actual store
process — after order confirmation — rather than being pre-determined sell-through. Shein takes this further: produces small test batches (100–
by production schedule. 200 units), reads demand within 7 days, then scales winners.

Result: ~30% reduction in finished goods inventory of wrong-color vehicles Result: Dramatically lower markdown rates vs. traditional fashion retailers
at Korean plants. (Zara ~15% vs. industry average 30–40%).

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University

Key Takeaways
BULLWHIP Demand variability amplifies upstream — small consumer fluctuations become large manufacturer swings. Post-
EFFECT COVID examples: semiconductor panic ordering, PS5 shortage, EV battery overcapacity. The BWE causes excess
inventory, poor service, and distorted production plans.

4 ROOT Demand forecast updating, order batching, price fluctuations, and rationing gaming — each requiring a different
CAUSES
countermeasure. Share POS data, use EDI, apply EDLP, and allocate based on historical sales.

RISK Aggregate demand across locations or products reduces variability: σ_pool = √(Σσᵢ²) < Σσᵢ. Always less than the
POOLING sum of individual σs. Enables same SL with less inventory, or better SL with same inventory. Benefit is largest
when streams are independent and similarly sized.

SERVICE Differentiate service levels by SKU: higher SL for high-margin, high-volume, low-variability, short-lead-time items.
LEVEL Applying uniform SL to all products wastes inventory where it adds little value and under-stocks where it matters
most.

POSTPONE- Delay the differentiating step in production until demand is known. Benetton (dye late), Apple (common chip),
MENT
Zara (central stock + rapid replenishment) — all exploit risk pooling through design and process choices.

Supply Chain Management | Prof. Seung Jun Lee | Chung-Ang University