Inventory classification tells a planner which items deserve attention and which do not. ABC, VED, FSN, and XYZ are the names most procurement and maintenance teams already know.
Most classification content stops at explaining what those four letters mean and how to calculate the thresholds. It rarely asks which dimension you should actually be classifying on, and what happens when demand shape, not value or criticality, is the variable that matters most.
This article is written for planners, SAP MM and PM users, and reliability engineers who already know what ABC, VED, FSN, and XYZ stand for, and who need the layer of judgment that decides which to reach for, when to combine them, and how to handle the one dimension almost no content treats seriously.
Type vs Class: The Distinction Most Inventory Content Collapses
Inventory type and inventory class get used interchangeably in most content, but they answer entirely different questions. Type describes what an item physically is, and it is fixed the moment the item enters the system.
Raw Material Unprocessed inputs to production | Work-in-Progress Partially completed production units | Finished Goods Ready for sale or shipment | MRO / Spares Maintenance, repair, and operating parts |
Consumables Used up in operation, not tracked as assets | Stock de sécurité Buffer held against demand or lead-time variability | In-Transit Owned, but physically moving between locations | Stock invendu No forecasted future use |
Class is different. Class is a behavioral tier derived from data: how often an item moves, what it costs, how consequential its absence is, and how predictable its demand looks when it occurs.
An item's type almost never changes. A spare part stays an MRO item for the entire time it sits in the storeroom. Its class is a different story: whether that same part sits in the A tier or the C tier this quarter depends on data that updates every time it is issued or superseded.
A mechanical seal is type MRO/spares from the day it is received. In its first year, low consumption places it in the C tier by value and the N tier by movement. Eighteen months later the asset it protects moves to continuous operation, and its class shifts to A/F while its type never moves at all.
That is the reason a classification exercise done once and filed away quietly goes stale, and it is the argument for everything that follows: classification is a live calculation, not a one-time label.
What Question Is Each Classification Framework Actually Answering?
Every classification framework answers a different business question. Choosing a framework is really choosing which risk you want to manage first.
Most organisations only ever answer one of these questions when they should be answering at least two.
| Framework | Question It Answers | Business Risk Managed |
|---|---|---|
| ABC | Which items concentrate the most capital? | Capital exposure |
| VED | Which items cause the most damage if they run out? | Stockout consequence |
| FSN | Which items move, and which sit still? | Obsolescence and storage risk |
| XYZ | How predictable is demand, in basic variability terms? | Forecast confidence |
| Syntetos-Boylan | What shape does demand actually take over time? | Forecasting method fit |
ABC is a value concentration exercise. It answers a capital question and nothing else, so an A-tier item by ABC alone tells you where the money sits, not whether the item matters operationally.
VED asks the opposite question. A vital item can be inexpensive, and a desirable item can be expensive, which is why ABC and VED are so often run together rather than in isolation.
FSN is a movement lens, useful for obsolescence risk and space allocation, but it says nothing about cost or consequence on its own. XYZ is the closest of the four classic frameworks to a demand-based view, though it typically works from a simple variability measure across all periods, coarser than the demand-pattern classification covered later in this article.
Your choice of framework should follow from which business risk you are managing, capital efficiency, stockout consequence, obsolescence, or forecast accuracy, not from habit or whichever field your ERP happens to have available.
The thresholds and calculation steps behind ABC, VED, FSN, and XYZ are covered in Verdantis's guide to applying the four classic spares classification models. This article focuses on which framework fits the risk you are actually managing, and the one framework most content leaves out entirely.
Single-Criterion vs Multi-Criterion Classification
A single classification lens fails for spare parts specifically, and it fails in a predictable direction. Value and criticality are frequently inverse.
A ten-dollar gasket can stop an entire production line. A five-thousand-dollar backup motor might sit unused for years without consequence. Multi-criterion classification exists to correct for exactly this mismatch, and two mechanics dominate in practice.
Here is what a 3×3 ABC-VED cross-tab looks like once built, with a distinct handling posture assigned to each of the nine cells.
| Essentiel | Indispensable | Souhaitable | |
|---|---|---|---|
| A | Tightest control Continuous review, highest safety stock | High control Frequent review | Managed control Watch for over-investment |
| B | High control Frequent review | Standard control Periodic review | Standard control Periodic review |
| C | Managed control Ensure availability despite low value | Light control Minimal review | Light control Reactive only |
Notice the C-Vital cell. It is the one every value-only classification misses: low cost, but essential enough that a stockout is unacceptable.
That single cell is usually where a pure ABC program quietly under-stocks the parts generating the most emergency purchase orders.
Matrix cross-tabs trade precision for a paper trail. Weighted indices trade the paper trail for precision. Choose the matrix when a planner or auditor needs to see why an item landed where it did.
Most classification programs never test for demand shape. Find out what your intermittent and lumpy items look like once they are separated out from the rest of the catalogue.
Reconnu par les entreprises du classement Fortune 500 et Global 2000
The Demand-Pattern Gap: Syntetos-Boylan Classification in Depth
ABC, VED, FSN, and XYZ all classify based on value, consequence, or movement. None of them classify based on the actual shape of demand over time.
That gap has a name: Syntetos-Boylan classification, and it is the framework most inventory content leaves out entirely, despite determining which forecasting method will actually work on a given part.
What ADI and CV² Actually Measure
Average Demand Interval (ADI) is the mean number of periods between two consecutive periods of non-zero demand. It is the total periods in the lookback window divided by the number of periods with any demand at all.
An ADI of 1 means demand occurs every period. An ADI of 6 means demand occurs, on average, once every six periods.
The squared coefficient of variation (CV²) measures the variability of demand size during periods when demand occurs, ignoring zero periods entirely. It is the squared ratio of the standard deviation of non-zero demand to its mean.
A low CV² means demand size is consistent whenever it happens. A high CV² means demand size swings wildly from one occurrence to the next.
The Four-Quadrant Split: Smooth, Erratic, Intermittent, Lumpy
Syntetos, Boylan, and Croston (2005) established cutoff values of 1.32 for ADI and 0.49 for CV², and these remain the values most widely cited in current forecasting research.
Erratic ADI < 1.32, CV² ≥ 0.49. Frequent, but size swings hard each time. Example: a specialty flange gasket ordered often, but in wildly different quantities each time. | Lumpy ADI ≥ 1.32, CV² ≥ 0.49. Infrequent and unpredictable in size. Hardest to forecast. Example: a large custom gearbox spare, ordered rarely and in unpredictable quantities. |
Smooth ADI < 1.32, CV² < 0.49. Frequent and consistent. Standard smoothing works fine here. Example: common O-ring seals consumed on a steady weekly schedule. | Intermittent ADI ≥ 1.32, CV² < 0.49. Infrequent, but consistent in size when it occurs. Example: an emergency shutdown valve actuator, rarely used but always replaced one at a time. |
Smooth demand behaves like the demand curves most forecasting tools are built for: frequent occurrence, consistent size. Lumpy demand sits at the opposite end of both axes, and it is where standard forecasting tools fail hardest.
Erratic and intermittent occupy the two off-diagonal quadrants. An erratic item needs a wider margin on size, while an intermittent item needs a wider margin on timing.
Here is how lopsided this actually gets. One published analysis of a global manufacturer's spare parts catalogue, a 2024 Erasmus University thesis applying the Syntetos-Boylan thresholds to real consumption data, found that 97.6% of items fell into the combined intermittent-or-lumpy category, with erratic at just 1.4% and smooth at only 0.9%.
| Smooth | 0.9% | |||
| Erratic | 1.4% | |||
| Intermittent + Lumpy | 97.6% |
That is not a minor tilt. A value-based or movement-based classification alone will spend most of its attention on a demand shape that barely exists in a typical spares population.
Why Standard Exponential Smoothing Fails on Intermittent Demand
Standard exponential smoothing updates its forecast every period, including zero-demand periods. For an intermittent item, that drags the forecast toward zero through every silent period, then produces a biased spike the moment real demand shows up.
The bias is not random noise that averages out. It is a structural property of applying period-by-period smoothing to a series that is mostly zeros, and it worsens the longer the silent stretches run.
Croston's method corrects this by updating only on non-zero periods. It forecasts demand size and demand interval as two separate series, then combines them into a single rate estimate.
The Syntetos-Boylan Approximation (SBA) then corrects a known upward bias in Croston's original formula. Conceptually, it shaves the Croston estimate down by a small, fixed percentage tied to the smoothing parameter. Written out, the relationship is shown below.
A further variant, the Teunter-Syntetos-Babai (TSB) method, updates demand probability at every period rather than only on non-zero ones, making it more responsive to genuinely declining demand and better suited to parts drifting toward obsolescence.
Classification and forecasting answer different questions
A classification framework describes what the demand data looks like. A forecasting method prescribes which algorithm to run given that shape. Treating a Syntetos-Boylan quadrant as a forecasting method, or a forecasting method as a classification tier, is one of the most common errors in existing inventory content.
The quadrant tells you the shape. Croston, SBA, and TSB are three different answers to what you do once you know the shape.
Data Mechanics That Make or Break Any Classification
The classification framework is only half the exercise. How the underlying data is prepared, cleaned, and windowed decides whether the resulting classification means anything at all.
Four data decisions that decide the result
Get these wrong and the framework choice above does not matter.
Lookback ABC and VED can run on 12 months of data. Demand-pattern classification needs 24 to 36 months for a stable ADI. | Zero-Demand Including or excluding zero-demand periods changes the ADI and CV² calculation. Fix the rule once, apply it to every SKU. | Seasonality Decide whether to adjust for seasonality before classifying, or let it register as inherent variability. Either can shift the quadrant. | New Item An item with no consumption history cannot be classified on demand at all. Use BOM linkage or criticality as a surrogate tier. |
Lookback window and zero-demand handling get skipped most often, usually because the same 12-month window and zero-inclusion rule that work fine for ABC get copy-pasted into a demand-pattern exercise unchecked.
A 12-month window against an item with an ADI near 4 only captures two or three demand events, not enough to distinguish a genuinely intermittent part from one that simply had a quiet year.
A part with a strong seasonal spike, deseasonalized incorrectly, can present as lumpy when it is actually smooth-but-seasonal. The two call for completely different stocking responses.
Static vs Dynamic Classification: When and How to Reclassify
A classification exercise run once and filed away is already going stale by the time anyone reviews it. Reclassification needs both triggers that say when to re-run the calculation and guardrails that stop the calculation from producing noise instead of signal.
That covers when and why to reclassify. What follows is a compact look at five more classification lenses worth knowing, before returning to industry context and ERP mechanics.
Five More Classification Lenses Worth Knowing
Value, criticality, movement, predictability, and demand shape are not the only lenses in play. Five more show up regularly in practice, and each answers a question none of the frameworks above can.
None of them gets the depth treatment this article gives demand-pattern classification, that depth belongs to the framework most content skips, but each is worth recognising by name so you reach for the right one instead of forcing the wrong tier to answer a question it was never built for.
Functional Classification: Why the Stock Exists
This lens asks why a given unit of on-hand stock exists, independent of what it is or how it behaves. Every quantity sitting in a storeroom is there for a specific operational reason, and that reason has a name.
Sizing these categories relies on a small set of standard formulas. In each, d is average demand per period, L is average lead time, sigma-d is the standard deviation of demand, sigma-L is the standard deviation of lead time, Z is the service-level factor, and Q is the order quantity.
Here is what that jump from 95% to 99% actually looks like in practice.
| 95% service level | 15 units | |||
| 99% service level | 29 units |
Ownership and Custody: Who Owns It, Who Reorders It
Two identical parts on the same shelf can belong to entirely different balance sheets, and to different job descriptions. This lens determines who carries the asset, who carries the risk, and who reorders it.
| Modèle | Balance Sheet | Reorder Duty |
|---|---|---|
| Company-owned | Full exposure on buyer's books | Buyer's planning team |
| Consignment | Asset stays on supplier's books until consumed | Buyer triggers use, supplier replenishes |
| VMI | Ownership terms vary by agreement | Supplier decides replenishment timing |
| Pooled / shared | Shared or allocated across sites | Shared governance across pool members |
Vendor-managed inventory and consignment get confused constantly, and the distinction has real commercial consequences. VMI is about who decides replenishment timing and quantity. Consignment is about who owns the stock until it is used, a distinction the Chartered Institute of Procurement and Supply lays out clearly.
You can have VMI without consignment, consignment without VMI, or both together in the same agreement, and treating them as synonyms leads to negotiating the wrong terms.
Physical and Handling Characteristics
This lens is supporting context rather than a strategic pillar, but it drives real decisions around storage design and safety compliance.
Perishable / Shelf-Life Requires rotation and expiry tracking, e.g. adhesives and sealants with a defined shelf life. | Hazardous Requires regulated storage and disposal, e.g. solvents and compressed gas cylinders needing MSDS documentation. | Oversized / Serialized Needs dedicated storage and unit-level traceability, e.g. a generator tracked by individual serial number. |
Obsolescence and Utilization Status
Active, excess, surplus, and obsolete inventory get used interchangeably, and they are not the same thing. The difference determines whether a write-down is optional or mandatory.
| Status | Définition | Example |
|---|---|---|
| Active | Consumed at a normal, forecastable rate | A bearing installed and running in a pump still in service |
| Excess | Exceeds projected demand but still has a demand path | A safety stock buffer grown larger than current usage justifies |
| Surplus | A subset of excess, typically from over-ordering | Duplicate safety stock of the same seal under two part numbers |
| Obsolete / Dead | No remaining demand path at standard value | A control card for a decommissioned asset, with no remaining use |
25 to 40% of MRO inventory at asset-heavy industrial sites is typically excess, obsolete, or duplicated. Catching a part while still excess, before it drifts into obsolete, is the difference between recovering some value and writing it off entirely.
Linking every part to the assets it serves via asset BOM management is often the clearest signal a part has lost its demand path, which is the mechanism behind systematic obsolescence detection rather than an annual audit.
Strategic Sourcing Risk: The Kraljic Matrix
Originally built for procurement category strategy, the Kraljic matrix classifies items on two axes, supply risk and operational impact, entirely independent of consumption value or failure consequence. CIPS maintains a detailed practitioner guide to scoring both axes from supplier count, substitute availability, lead time exposure, spend, and output impact.
Leverage Low risk, high impact. Example: bulk lubricants purchased from many qualified suppliers at meaningful spend. | Strategic High risk, high impact. Example: a custom-engineered, sole-sourced turbine rotor. |
Non-Critical Low risk, low impact. Example: standard fasteners available from dozens of distributors at low spend. | Bottleneck High risk, low impact. Example: a specialty sensor available from a single overseas supplier, at low spend. |
Kraljic classifies supply risk. ABC classifies consumption value. Neither classifies failure impact on its own, which is why a part can be cheap and easy to source and still be mission-critical if its failure stops a production line.
Confusing these questions is how a genuinely critical spare ends up under-stocked. Failure-impact prioritization for critical spares is a separate, complementary lens to sourcing risk, not a substitute for it.
Context Changes Which Dimension Should Dominate
The right classification dimension depends on which industry, and which decision, is sitting on the other end of it. There is no universally correct weighting between value, criticality, movement, and demand shape.
The same part number is not the same risk everywhere
A bearing that is routine and easily substituted at one plant can be a single point of failure at another, simply because of which asset it supports. Classifying a part number once, at the enterprise level, is one of the least visible ways multi-site programs get individual sites wrong.
Where Classification Actually Lives in Your ERP
Most classification tiers never make it past a spreadsheet, even when the ERP has a field built for exactly this purpose, because the field, the ownership, and the update trigger were never connected.
Material master ABC indicatorThe native field most planners default to, typically on the material master's plant data view. Handles value-based tiers well but has no native concept of criticality, so it cannot carry a composite score alone. | Custom Z-fieldA common workaround when the native indicator cannot hold a composite score. Flexible to configure, but invisible to anyone who does not already know the Z-field exists. | PM-side criticality indicatorCriticality belongs on the equipment or functional location, then inherits down to every part on that asset's bill of materials, rather than being assigned directly and inconsistently at the part level. |
Criticality inheritance runs through the equipment's bill of materials, which is why assigning criticality directly to a part instead of the asset produces inconsistent tiers across identical parts.
The second gap is procedural, not technical: whether a tier shift from C to A automatically triggers an MRP type or reorder policy change, or sits unnoticed until a planner intervenes manually. The second scenario is far more common, and it is the single biggest reason classification programs quietly stop producing value after the first year.
| Approach | How It Runs | Operational Implication |
|---|---|---|
| Stored / periodic | Recalculated on a schedule and written to a field. | Predictable and auditable, but can lag a real demand shift by weeks. |
| Live-query | Calculated on demand when a report or MRP run needs it. | Always current, but harder to audit and more expensive at scale. |
From Classification to Decision: Where Most Content Stops Short
Classification is an input, not a deliverable. A tier that does not change what a planner actually does next is not doing any real work, no matter how carefully it was calculated.
| Tier | Review Frequency | Safety Stock Method | Reorder Policy | Vendor Management |
|---|---|---|---|---|
| A / High | Monthly | Service-level driven, demand-pattern aware | Continuous review | Named backup supplier, expedite terms pre-negotiated |
| B / Medium | Quarterly | Fixed buffer with periodic adjustment | Periodic review | Standard supplier terms, reviewed at renewal |
| C / Low | Annually | Minimal or none | Reactive / on-demand | No dedicated management, ordered as needed |
Use this as the closing test for any classification program: if two tiers do not produce two different answers on review frequency, safety stock method, or reorder logic, the classification is not doing real work.
This tier-to-lever mapping is exactly the layer covered in Verdantis's guide to MRO inventory management and stocking policy, for readers whose next step is building the policy side.
Frequently Asked Questions
What is the difference between inventory type and inventory class?
Type describes what an item physically is, like raw material or MRO spares, and it is fixed. Class is a data-derived behavioral tier, like an ABC rank, that can change even though the item's type never does.
Do I need to run every classification framework on every item?
No. Pick the framework, or combination, that matches the risk you are managing: ABC for capital exposure, VED for stockout consequence, FSN for movement, XYZ for basic predictability, and Syntetos-Boylan for demand shape.
How is Syntetos-Boylan classification different from ABC or XYZ?
ABC classifies by value and XYZ by a general variability measure. Syntetos-Boylan classifies specifically by the shape of demand over time, using ADI and CV², a distinction most other frameworks do not attempt.
What are the standard ADI and CV² cutoff values?
Syntetos, Boylan, and Croston (2005) established cutoffs of ADI = 1.32 and CV² = 0.49, and these remain the values most widely cited in current forecasting research.
How much demand history do I need for demand-pattern classification?
Plan for 24 to 36 months. Intermittent items need enough non-zero demand cycles to produce a stable ADI and CV² calculation, and a 12-month window is usually too short.
Should I classify at the SKU level or the SKU-per-location level?
SKU-per-location, wherever the stocking decision actually gets made. A global SKU classification can hide a part that behaves smoothly at one plant and erratically once the network is aggregated.
What causes reclassification churn, and how do I stop it?
Items sitting near a threshold oscillate between tiers from ordinary noise, not real behavior change. A hysteresis buffer around each threshold prevents these false reclassification events.
Where should criticality live in an ERP, on the part or the equipment?
On the equipment or functional location. Criticality should inherit down to every part on that asset's bill of materials, rather than being assigned directly and inconsistently at the part level.
Does a tier change automatically update the reorder policy in most ERPs?
Rarely, by default. Most systems require a planner to manually update the MRP type or reorder policy after a tier shift, a common and often unnoticed implementation gap.
How do I classify a brand-new item with no consumption history?
Use a surrogate classification based on BOM linkage or equipment criticality until enough real consumption data accumulates to support a genuine demand-based tier.
What is the difference between Croston's method and the Syntetos-Boylan Approximation?
Croston's method forecasts intermittent demand by updating only on non-zero periods. SBA applies a bias correction to Croston's original formula, which is known to systematically overstate demand.
How often should inventory classification be reviewed?
Tie review frequency to the tier itself. High-criticality or high-value tiers warrant more frequent review, and any BOM change or plant relocation should trigger an off-cycle review regardless of schedule.
What is decoupling stock, and how is it different from safety stock?
Decoupling stock sits between two production stages so a short upstream stoppage does not halt downstream work. Safety stock buffers against demand or lead-time variability from an external supplier.
Is vendor-managed inventory the same as consignment inventory?
No. VMI describes who controls replenishment timing and quantity. Consignment describes who owns the stock until it is used. The two can exist independently or together in the same agreement.
What's the difference between excess, surplus, and obsolete inventory?
Excess inventory still has a demand path, just more than currently needed. Surplus is a subset of excess, typically from over-ordering. Obsolete inventory has no remaining demand path.
How is the Kraljic matrix different from a criticality assessment?
Kraljic measures supply risk and sourcing exposure. Criticality measures the operational impact if a part fails. A part can score low on one and high on the other.
See these classification models in action
Read how Fortune 500 and Global 2000 manufacturers apply criticality and demand-pattern classification across real MRO and MDM programs.
Read the case studies
