Week 2 | Session 5: Inventory Segmentation - Methods, Advanced Approaches & AHP Case

Course: Supply Chain Digitization

---

Recap & Session Context

Note

Session 4 covered: Analytical product segmentation using the Demand Variability vs. Weekly Sales quadrant (CoV-based) → 4-quadrant matrix → Push / Pull / Hybrid strategies, and the Kraljic Matrix for procurement.

This session: Inventory Segmentation - what it is, the four popular single-criterion methods, five advanced multi-criteria approaches, and a full AHP-based case study.

This is the final session of Module 2 - Supply Chain Segmentation.

---

What is Inventory Segmentation?

Note

Definition: Categorising a company’s inventory into distinct groups based on defined criteria, so that each group receives a tailored management strategy.

Key benefits:

Benefit	Description
Cost optimisation	Allocate SC resources precisely where they generate the most value
Improved service levels	Set and maintain the right service level per SKU category
Right storage system	Assign appropriate storage infrastructure per inventory type
Better picking strategy	Design warehouse and DC picking processes based on movement frequency and item characteristics
Operational performance	Overall improvement across fulfilment, replenishment, and inventory accuracy

---

Popular Inventory Segmentation Methods - 4 Traditional Approaches

Note

All four traditional methods are single-criterion approaches. While each is powerful within its domain, real business decisions require multiple criteria simultaneously. Single-criterion methods risk oversimplification and suboptimal strategies - which is why advanced multi-criteria methods exist.

1. ABC Analysis - Revenue-Based

Criterion: Revenue Contribution

Classification based on the revenue contribution of each inventory item - the Pareto principle applied to inventory.

Segment	Share of Products	Share of Revenue	Management Priority
A	~20%	~80%	Highest - tightly monitored, frequent review
B	~30%	~15%	Moderate - regular review
C	~50%	~5%	Lowest - simplified, automated management

Note

★ ABC = Pareto principle applied to inventory. 20% of SKUs drive 80% of revenue - focus management effort accordingly.

---

2. FSN Analysis - Movement-Based

Criterion: Speed of Movement

Classification based on the consumption rate and speed of movement through the warehouse.

Segment	Share of Items	Movement Behaviour	Avg Cumulative Stay
F - Fast Moving	~10%	Very short stay in warehouse; consumed rapidly	Less than 10% of average cumulative stay
S - Slow Moving	~30%	Moderate movement; stays longer in warehouse	~20% of average cumulative stay
N - Non-Moving	~50%	Stagnant inventory - no consumption for extended period; inventory turnover ratio < 1	High cumulative stay

Tip

Application of Concepts: Non-Moving (N) items are a direct holding cost liability. Identifying them allows the business to decide whether to liquidate, discount, or discontinue - rather than continuing to pay storage costs on stock that will never move.

---

3. VED Analysis - Criticality-Based

Criterion: Operational Criticality

Classification based on the criticality of each item to business operations - not its value or movement speed.

Segment	Meaning	Description
V - Vital	Cannot operate without it	Absolutely crucial - business halts if unavailable
E - Essential	Very important	High priority after Vital; significant disruption if unavailable
D - Desirable	Not strictly necessary	Operations can continue without it, but quality or efficiency suffers

Note

★ VED is most common in manufacturing, hospitals, and defence - wherever certain items are mission-critical regardless of their cost or demand frequency. A low-cost item can still be Vital.

---

4. XYZ Analysis - Demand Variability-Based

Criterion: Demand Predictability

Classification based on the predictability of demand over time - complementary to ABC analysis.

Segment	Demand Behaviour	Management Approach
X	Little or no variation - highly predictable	Lean inventory; tight replenishment cycles
Y	Unsteady demand - but can be predicted to a certain extent	Moderate safety stock; regular review
Z	Very high variation - no discernible trend or causal factors	High safety stock; frequent monitoring; hardest to manage

---

Quick Reference - 4 Traditional Methods

Method	Classification Criterion	Segments
ABC	Revenue contribution	A (high) / B (medium) / C (low)
FSN	Speed of movement / consumption rate	F (fast) / S (slow) / N (non-moving)
VED	Criticality to operations	V (vital) / E (essential) / D (desirable)
XYZ	Demand predictability / variability	X (stable) / Y (variable) / Z (unpredictable)

---

Advanced Inventory Segmentation Approaches

Multi-Criteria Methods

Note

Enabled by: the combination of data availability and increased computational power - both now accessible to most organisations. Advanced methods handle two or more criteria simultaneously for realistic, business-grade segmentation.

1. Mathematical Programming

• Linear Programming (LP) or Non-Linear Programming (NLP)
• Formulates the segmentation problem as a mathematical optimisation model incorporating multiple criteria simultaneously

2. Metaheuristics

Used when the problem is too complex for exact mathematical programming:

Algorithm	Type
Genetic Algorithm (GA)	Evolutionary optimisation
Particle Swarm Optimisation (PSO)	Swarm intelligence
Simulated Annealing	Probabilistic local search

3. AI / Machine Learning

Heavily data-driven - well suited to today’s data-rich SC environment:

Algorithm	Type
Artificial Neural Networks (ANN)	Deep learning
Support Vector Machines (SVM)	Supervised classification
Back Propagation Networks	Neural network training
K-Nearest Neighbor (KNN)	Instance-based learning
Regression models	Predictive modelling

4. Multi-Criteria Decision Making (MCDM)

Incorporates expert opinion to weigh multiple factors:

Method	Description
AHP - Analytical Hierarchy Process	Pairwise comparison of criteria to derive priority weights
Fuzzy AHP	AHP with fuzzy logic to handle uncertainty in expert judgement
ANP - Analytical Network Process	Extension of AHP allowing interdependencies between criteria

5. Hybrid Approaches

Combinations of the above - e.g., AHP + ML, GA + LP - used when no single method is adequate alone.

Tip

★ Exam tip - Advanced methods: Know all four traditional methods in full detail. For advanced methods, be able to name each category and state why it is used: handling multiple criteria simultaneously, accommodating data availability, or managing complexity beyond single-criterion approaches.

---

AHP - Analytical Hierarchy Process

MCDM Framework

Note

Proposed by: Thomas L. Saaty

Type: Multi-Criteria Decision Making (MCDM) method

Applications: Supplier selection, procurement decisions, product and service design, resource allocation, and inventory segmentation.

Core Idea

AHP organises a complex, multi-factor decision into a hierarchical structure, performs pairwise comparisons of all factors to determine their relative importance, and outputs quantitative, consistent priority weights for each factor. Inputs can come from a single expert or aggregated from multiple experts.

AHP Hierarchy Structure

Define the problem & determine the objective

↓

Develop the hierarchical framework

↓

Construct pairwise comparison matrix for each level

↓

Calculate vector priorities & weights

↓

Calculate the Consistency Ratio (CR)

↓

Is CR < 0.1?

↓  yes

Use the obtained criterion weights

If CR ≥ 0.1 (No) → return to the pairwise comparison matrix and redo the judgements.

The AHP method hierarchically organises a complex problem, compares elements pairwise, and iterates until the judgements are consistent (CR < 0.1) before the derived weights are used.

Saaty Scale (1-9)

Note

Scale Value	Meaning
1	Equally important
3	Moderately more important
5	Strongly more important
7	Very strongly more important
9	Extremely more important
2, 4, 6, 8	Intermediate values between adjacent judgements

AHP Steps

1. Define the problem - state the objective clearly
2. Develop the hierarchical framework - list all criteria and map their relationships
3. Construct the Pairwise Comparison Matrix for each level using the Saaty scale (1-9)
4. Normalise the matrix - calculate criterion weights (the priority vector)
5. Calculate the Consistency Ratio (CR) using: CR = CI / RI where CI = Consistency Index and RI = Random Index (standard table value based on matrix size)
6. Check CR: If CR < 0.1 → consistent, proceed. If CR ≥ 0.1 → inconsistent, return to Step 3 and redo the pairwise comparison

Note

★ CR < 0.1 is the MUST-PASS condition in AHP. If this threshold is not met, the expert’s pairwise judgements are considered internally inconsistent and unreliable - the comparison must be redone.

---

Case Study - AHP-Based Inventory Segmentation: XYZ E-tailer

Case Study

Case Setup

Parameter	Detail
Company	XYZ E-tail - an e-commerce retailer
Constraint	Fixed warehouse space under a 3-year contract
Listing policy	A product is listed as ‘available’ only if it is physically present in the warehouse
Challenge	Management wants to expand the product portfolio but warehouse space is limited
Previous policy	95% service level maintained uniformly for all SKUs
New plan	Classify 25 SKUs into 3 groups with differentiated service levels
Tool	AHP - to classify 25 SKUs into Class A / B / C

Target service levels under the new classification:

Class	Service Level
A	95%
B	90%
C	85%

---

SKU Data - 25 SKUs Across 6 Criteria

Table 2 - SKU data across the six criteria:

SKU	Monthly Demand	Category Priority (1-10)	Supplier Reliability (%)	Profit Margin (%)	Lead Time (hrs)	Return Likelihood (%)
1	1,366,987	10	97	10	72	4
2	321	3	93	16	60	2
3	1,488,630	10	97	14	48	2
4	629	4	89	15	107	7
5	1,428,018	9	98	9	90	2
6	88,232	7	94	7	78	3
7	8,778	4	90	5	99	1
8	1,384,335	10	95	13	48	1
9	7,202	3	95	7	49	2
10	38	3	96	7	66	1
11	53	1	98	5	105	1
12	5,990	3	95	9	78	2
13	1,277	3	89	4	107	3
14	872,270	7	95	7	107	2
15	783,381	7	85	8	112	3
16	78	1	89	12	62	8
17	301	2	85	12	100	7
18	105	2	86	16	86	5
19	12	1	93	7	119	7
20	29	1	91	5	85	6
21	74	2	88	7	78	4
22	55	3	96	7	91	6
23	109	1	88	11	105	6
24	578,358	6	96	14	107	3
25	200	2	85	5	93	6

---

6 Evaluation Criteria

The six criteria used, with their direction of preference:

#	Criterion	Unit	Direction
1	Monthly Demand	Units/month	Higher = Better
2	Priority of Product Category	Score 1-10	Higher = Better
3	Supplier Reliability	% perfect orders	Higher = Better
4	Profit Margin	%	Higher = Better
5	Lead Time	Hours	Lower = Better
6	Likelihood of Return	%	Lower = Better

Note

★ Mixed directions matter: Criteria 5 and 6 are “Lower = Better.” Normalisation must account for this - they are inverted before scoring. Cannot compare or multiply criteria in different units or directions without normalisation first.

---

Step 1 - Pairwise Comparison Matrix

Table 1 - relative priorities of factors (1 = equally important; 9 = extremely more important):

	Monthly Demand	Category Priority	Supplier Reliability	Profit Margin	Lead Time	Return Likelihood
Monthly Demand	1	3	5	2	6	5
Category Priority	1/3	1	2	3	3	5
Supplier Reliability	1/5	1/2	1	2	4	5
Profit Margin	1/2	1/3	1/2	1	3	3
Lead Time	1/6	1/3	1/4	1/3	1	3
Return Likelihood	1/5	1/5	1/5	1/3	1/3	1

Example reading: Monthly Demand vs. Priority of Product Category → Monthly Demand rated 3× more important (Saaty scale = 3).

---

Step 2 - Normalised Matrix & Criterion Weights

Each column of Table 1 is divided by its column sum; the row averages give the criterion weights (priority vector):

Criterion	MD	PPC	ASR	PM	LT	LR	Weight
Monthly Demand	0.42	0.56	0.56	0.23	0.35	0.23	0.39
Category Priority	0.14	0.19	0.22	0.35	0.17	0.23	0.22
Supplier Reliability	0.08	0.09	0.11	0.23	0.23	0.23	0.16
Profit Margin	0.21	0.06	0.06	0.12	0.17	0.14	0.13
Lead Time	0.07	0.06	0.03	0.04	0.06	0.14	0.07
Return Likelihood	0.08	0.04	0.02	0.04	0.02	0.05	0.04

Step 3 - Consistency Check

Note

Result: CR = 0.08 < 0.1 → Consistent ✓

The expert’s pairwise comparisons pass the AHP consistency threshold. The derived criterion weights are valid and can be used for SKU scoring.

---

Criterion Weights Output

Criterion weights (priority vector)

• Monthly Demand - 0.39
• Category Priority - 0.22
• Supplier Reliability - 0.16
• Profit Margin - 0.13
• Lead Time - 0.07
• Return Likelihood - 0.04

Consistency check

• λ_max = 6.51 · n = 6
• Consistency Index (CI) = 0.101
• Random Index (RI) = 1.24
• CR = CI / RI = 0.082
• CR < 0.1 → the matrix is reasonably consistent ✓

---

Normalisation of SKU Data

Problem: All six criteria are in different units - they cannot be directly compared or multiplied.

Solution: Normalise all criteria to a 0-1 scale before scoring.

Note

For Higher-is-Better criteria (Monthly Demand, Category Priority, Supplier Reliability, Profit Margin):

Normalised Score = (xᵢ − min) / (max − min)

For Lower-is-Better criteria (Lead Time, Likelihood of Return):

Normalised Score = 1 − [(xᵢ − min) / (max − min)]

This inversion converts “lower is better” into a consistent “higher is better” scale - ensuring all six criteria point in the same direction before being multiplied by their AHP weights.

Note

Each of the 25 SKUs’ six raw values is normalised to a 0-1 scale using the formulas above (with Lead Time and Return Likelihood inverted, since lower is better). Every normalised value is then multiplied by its criterion weight and summed to give one combined SKU Score per SKU - the scores used in the classification table below.

---

Calculating the SKU Score

Note

For each SKU, multiply each normalised criterion score by its AHP criterion weight, then sum across all six criteria:

SKU Score = Σ (AHP Weightᵢ × Normalised Scoreᵢ)

A higher combined score indicates a more strategically important SKU overall.

---

ABC Classification from SKU Scores

1. Sort all 25 SKUs by combined SKU Score in descending order (highest → most important)
2. Calculate cumulative score and the cumulative % of total score for each SKU
3. Apply cut-offs to assign classes:
   • Class A → top SKUs up to ~60% of cumulative score → 7 out of 25 items → 95% service level
   • Class B → next ~25% of cumulative score → 8 out of 25 items → 90% service level
   • Class C → remaining ~15% → 10 out of 25 items → 85% service level

Note

★ The cut-off percentages (60% / 25% / 15%) are not universal fixed values. They are set by the company based on its own business strategy, warehouse capacity constraints, and service level targets. A different company with different constraints would set different cut-offs.

---

Final Classification Results

SKUs sorted by combined score; cut-offs at ~60% / 85% of cumulative score assign classes A / B / C:

SKU	Score	Cumulative	% Cumulative	Class
3	0.96	0.96	10.7%	A
8	0.90	1.86	20.7%	A
1	0.85	2.72	30.2%	A
5	0.84	3.56	39.6%	A
14	0.58	4.14	45.9%	A
24	0.55	4.69	52.1%	A
15	0.43	5.12	56.8%	A
6	0.38	5.49	61.0%	B
2	0.36	5.86	65.1%	B
10	0.31	6.16	68.5%	B
9	0.31	6.47	71.9%	B
12	0.30	6.77	75.2%	B
22	0.25	7.02	78.0%	B
4	0.25	7.28	80.8%	B
11	0.23	7.51	83.4%	B
18	0.21	7.71	85.7%	C
7	0.21	7.92	88.0%	C
16	0.19	8.11	90.1%	C
21	0.15	8.26	91.8%	C
13	0.14	8.40	93.3%	C
19	0.14	8.54	94.8%	C
23	0.14	8.67	96.3%	C
17	0.13	8.80	97.8%	C
20	0.13	8.93	99.2%	C
25	0.07	9.00	100.0%	C

ABC classification using AHP

↓

A class28% of SKUs · 95% service level

B class32% of SKUs · 90% service level

C class40% of SKUs · 85% service level

Final result - the AHP score drives a differentiated service-level policy: class A (60% of total score) held at 95%, class B (25%) at 90%, class C (15%) at 85%.

---

Module 2 Summary - Supply Chain Segmentation (All 5 Sessions)

Session	Topic Covered
Session 1	SC challenges - building the case for why segmentation is needed
Session 2	8 reasons for segmentation + 7 types of segmentation
Session 3	Functional vs. Innovative products → Efficient vs. Responsive SC → Push / Pull / Hybrid → Push-Pull Boundary
Session 4	Analytical product segmentation (CoV quadrant) + Kraljic Matrix
Session 5	Inventory segmentation - ABC / FSN / VED / XYZ + Advanced methods + AHP multi-criteria classification case

Tip

★ Exam Tip - Inventory Segmentation: For traditional methods, always state the criterion used: Revenue = ABC | Movement = FSN | Criticality = VED | Variability = XYZ

For AHP questions: State all six steps clearly and always mention the CR < 0.1 condition - it is the threshold that validates the expert’s pairwise judgements.

Key conceptual contribution of this session: AHP provides a principled, data-driven way to handle multi-criteria, mixed-unit, and mixed-direction factors - combining them into a single consistent classification score. This is what traditional single-criterion methods cannot do.