The ABC-XYZ matrix crosses the two segmentations of this cluster: a reference’s financial weight and the regularity of its demand. These two dimensions are largely independent, and crossing them produces nine distinct steering regimes. Its value is not descriptive but prescriptive: each cell calls for its own replenishment policy, service level and forecasting effort.
Why it matters
The two preceding pages each answer a distinct question. Classification by value says where the financial stakes lie; regularity analysis says what can be expected of a forecast. Taken separately, both lead to incomplete recommendations, because each ignores half the problem.
Crossing them changes the nature of the exercise. A high-value but perfectly regular reference and a high-value but wholly erratic one belong to the same financial class and yet have almost nothing in common operationally. The first is steered finely and rewards every optimisation effort; the second resists any model and demands structural answers.
The signature above makes this independence visible. Each dot is a reference, positioned by financial weight along the columns and by regularity along the rows. The dots do not line up neatly on a diagonal: you find important and unstable items as well as negligible and perfectly regular ones. It is precisely this dispersion that justifies the matrix.
The practical benefit is to turn a portfolio of several thousand lines into nine homogeneous populations, each steerable by an explicit rule. Decisions stop being taken reference by reference, which is impossible at scale, without becoming uniform, which would be costly.
The cells carry neither the same weight nor the same interest. Walk through them in the signature: only a few references occupy the high-value, high-regularity cell, and yet they concentrate most of the capital committed. That is where fine optimisation returns the most per unit of effort, because regularity makes every gain durable and predictable.
At the opposite corner, the erratic long tail often gathers a third of the lines for a few percent of the value. The classic error is to devote considerable time to it, precisely because these references cause visible daily problems. The right answer is to install simple, robust rules there, accept lower service, and sometimes raise the question of rationalising the catalogue.
The most interesting cell remains the one combining high value with low regularity. It is sparsely populated, difficult, and it escapes the usual answers: no forecasting improvement will save it, no reasonable buffer will cover it. The levers that work there are structural, decoupling, lead-time reduction, pooling across sites or a move to make-to-order. A well-built matrix serves above all to isolate that handful of references and give them the leadership attention they deserve.
The mechanism
Construction is mechanical once both rankings exist. You assign each reference its value letter and its regularity letter, and obtain its cell. The useful work only begins afterwards: writing, for each of the nine cells, the steering rule that applies by default.
That rule contains at least four elements. The target service level, which sets buffer thickness. The replenishment policy, continuous monitoring or periodic review. The forecasting method, from a carefully built model to a simple average. And the mode of human handling, individual follow-up by a planner or fully automatic management with alerts by exception.
Reading the columns and the rows follows two different logics, and both must be held together. Going down a column, value stays constant and regularity degrades: the buffer thickens and forecasting loses its power. Moving along a row, regularity stays constant and value decreases: steering effort must fall, not because the reference is easier, but because it deserves less attention.
Finally, the matrix is revised, and less often than one might think. Both rankings evolve slowly, and a quarterly or half-yearly revision suffices in most contexts. What must be handled continuously, however, are cell changes: a reference moving from one cell to another changes steering regime, and that shift deserves to be flagged rather than absorbed silently.
The traps
A matrix displaying nine headcounts and no policy has produced nothing. The value of the exercise lies entirely in the rules attached to each cell, target service, policy, forecasting method and follow-up mode. Without them the grid is one more report.
Low-value, irregular references generate many visible incidents and therefore attract attention. Devoting time proportional to the noise they make, rather than to the value they carry, diverts effort from the cells where it would pay.
A reference switching regime is often still steered by its old parameters, for want of an alert. This is particularly costly when an item gains value while continuing to be treated as a secondary reference.
The rollout
Establish the value classification and the regularity analysis separately, on clean history and at coherent buckets, before any crossing.
Document, for each of the nine cells, the service level, the replenishment policy, the forecasting method and the follow-up mode. That document is the real deliverable.
Isolate high-value, low-regularity references and apply structural levers to them, decoupling, lead-time reduction, pooling, rather than hoping for a gain from parameter settings.
Put simple, robust rules on the low-value cells, managed by exception, and explicitly raise the question of keeping references that carry neither value nor regularity.
Recompute the matrix at regular intervals and raise an alert on every regime change, so that parameters actually follow the reference’s new status.
Neighboring concepts
From knowledge to action
That handful of references resists parameter tuning and calls for structural levers. Our Inventory & distribution file builds the matrix and writes the steering rule for every cell.