Segmentation Is Dead. Long Live Segmentation

Why Your ABC-XYZ Analysis Is Quietly Mislabeling Your Best Products

ABC-XYZ analysis has been a cornerstone of demand planning for nearly a century. ABC classification ranks products by revenue contribution. XYZ attempts to classify demand variability using a single, familiar metric — the Coefficient of Variation (CV). It is quick, intuitive, and widely taught.

It is also quietly getting some of your most forecastable products wrong.

At Valtitude, we work with planning teams across industries to sharpen demand segmentation and forecasting accuracy, and one pattern shows up consistently: the CV-based XYZ classification systematically mislabels products with strong seasonal patterns and clear trends as "erratic" — placing them in the X or Z bucket, where they receive safety stock buffers and manual overrides they don't need.

The result is a segmentation scheme that overcomplicates the forecast and underserves the genuinely unpredictable.

What CV Measures — and What It Doesn't

The Coefficient of Variation is calculated as the ratio of the standard deviation to the mean. It is a compact, back-of-the-envelope measure of relative variability. Its appeal is obvious: one number, easy to compute, easy to rank.

The problem is what that single number contains. CV lumps together every source of movement in a time series into one figure:

• Trend — the long-run upward or downward direction of demand

• Seasonality — recurring patterns tied to calendar effects

• Cyclical variation — medium-term demand cycles

• True randomness — the residual noise that genuinely resists forecasting

A product with a strong seasonal pattern and a clean underlying trend can post a CV well above 50% and land in the "unforecastable" bucket — even though a basic time series decomposition would fit it almost perfectly. The CV is high not because the product is unpredictable, but because it moves a lot in ways that are entirely predictable.

The Classic Illustration: Airline Passengers

Consider the well-known airline passenger time series — a benchmark dataset used in forecasting research for decades. Its raw CV is approximately 43%, which under a standard XYZ scheme would classify it as highly variable, borderline erratic.

Decompose that series, however, and the picture changes completely. Roughly 91% of the variation is explained by trend, approximately 9% by seasonality, and less than 1% remains as unexplained residual. That is not a Z product. That is a highly forecastable product wearing the wrong label — and being managed accordingly.

A More Honest Segmentation Criterion: Residual CV

The alternative we advocate is straightforward in concept: compute the CV not on the raw demand series, but on what remains after trend and seasonality have been stripped out. We call this the Residual Coefficient of Variation, or RCV — a measure of the unforecastable ratio in your demand signal.

RCV answers the question your segmentation should be asking: After accounting for everything a statistical model can explain, how much genuine uncertainty is left?

Two SKUs with identical raw CVs can have dramatically different RCVs. One you can forecast to within a few percentage points with a well-configured model. The other genuinely resists prediction regardless of the algorithm applied. Treating them the same — stocking them the same, reviewing them with the same frequency, applying the same planning policies — is precisely how demand planners end up over-buffering the predictable and under-investing in the products that actually require planner judgment and safety stock discipline.

What This Means for Your Planning Process

Shifting from CV to RCV as your primary segmentation criterion has practical downstream effects across your planning stack:

• Forecasting model selection becomes more precise. Products with low RCV are well-suited to statistical models — Holt-Winters, exponential smoothing, or trend-seasonal decomposition. Products with high RCV are candidates for judgment-heavy planning, Demand Sensing, or collaborative input from sales and marketing.

• Safety stock policy becomes more defensible. Safety stock is designed to buffer against uncertainty. If your segmentation inflates the apparent uncertainty of forecastable products, your safety stock calculations will systematically over-invest in the wrong places.

• Planner time and attention are a finite resource. Exception-based reviews should concentrate on products that genuinely cannot be modeled — not on seasonal items that look volatile when viewed through a CV lens.

The Right Question for Demand Planning Leaders

The debate is not whether ABC-XYZ analysis is useful — it is. The debate is whether the XYZ criterion you are using measures what you intend to measure.

CV measures total variability. RCV measures unexplained variability. For demand planning, the second question is the one that matters.

If your current segmentation is based purely on CV, there is a reasonable chance that a portion of your "erratic" portfolio is, in fact, quite forecastable — and that the planning policies applied to those products are working against you.

Segment on what is unexplainable. Not on what merely moves.

Rethinking Demand Variability Measurement:

ABC-XYZ segmentation remains a powerful planning framework — but its default metric, the Coefficient of Variation, conflates forecastable movement with genuine uncertainty. By decomposing demand into its structural components and measuring what truly resists forecasting, the Residual CV gives demand planners a sharper, more actionable lens for segmentation, safety stock policy, and model selection.

At Valtitude, we help planning teams move beyond default configurations and build segmentation logic that reflects the real structure of their demand. If your team is ready to take a closer look at what your XYZ classification is measuring — and what it might be missing — let's start that conversation.

Contact US