Stop Segmenting. Start Deciding.
For years I’ve watched well‑intentioned teams split their assortments, customers, suppliers, and stores into tidy little boxes—A, B, C; Gold, Silver, Bronze; “strategic” versus “non‑strategic.” The promise is focus and clarity. The result is almost always the opposite: more moving parts, more meetings, and less control over what actually matters—profitable decisions under uncertainty. I make this argument at length in Introduction to Supply Chain; consider this piece a field note on why the segmentation instinct is so persistent—and why it’s time to outgrow it.
The lure—and trap—of segmentation
Segmentation was born in a world of paper ledgers and scarce compute. Before cheap, pervasive inventory systems, the A/B/C hierarchy helped reduce clerical work: treat a few “A” items tightly, let the “C” items fend for themselves. That world is gone. What remains is the habit. In modern practice, ABC is a crude categorization that survives mostly as a reporting device and as a way to parcel out human attention—pick thresholds, color some dashboards, and feel in control. The parameters that define those classes are convenient rather than principled; they tend to be rounded for memorability, not derived from economics.
Once you start slicing, the slices take over. KPIs, reviews, and policies get copied per segment. Your organization now runs several miniature supply chains in parallel, each with its own rituals. The irony is that the supposed simplification increases the surface area you must manage. In practice, the segmentation becomes part of the problem space—an object to maintain—rather than a means to better decisions.
Worse, the picture is low‑resolution. ABC ignores anything dynamic; ABC‑XYZ adds variance but still misses trends, launches, and seasonality. Two items traveling in opposite directions through time can land in the same cell and inherit the same policy. When reality moves, your categories don’t. The “insight” was an artifact of the grid.
And then comes the churn. Seemingly minor changes in the window or thresholds reshuffle items across borders. In many contexts, a meaningful share of the catalog flips class quarter after quarter. Policies tied to classes flip with them, creating discontinuities in service and stock—precisely the kind of erratic behavior we should be trying to avoid.
Finally, segmentation blurs money. Frequency is not importance. A rarely sold spare part can ground an aircraft; a low‑turn “window” item can enable baskets across a store. Yet class‑based policies treat such items as unimportant by construction, and ABC‑XYZ formalizes this myopia with rules that look rigorous while remaining blind to indirect effects. This is cargo‑cult mathematics—moments and matrices that confer a false sense of precision without measuring what the business actually optimizes.
If this all feels familiar, it is because segmentation is a heuristic—simple, memorizable, reassuring. Heuristics are not evil; they are just unreliable when elevated to doctrine. In supply chains, where choices are discrete and numeric rather than instinctive, rules of thumb that “feel right” often can’t withstand basic financial scrutiny.
The hidden bill
Segmentation looks cheap. It is not.
It creates instability at the borders—policy whiplash when items “jump” classes. It creates bureaucracy—meetings to argue thresholds, variants to patch edge cases, and entire dashboards whose only job is to explain the last reclassification. It creates distraction—bikeshedding around parameters that are easy to debate precisely because they are arbitrary. Every hour spent litigating whether 10% or 12% defines “Y” is an hour not spent improving the economics of a decision.
It also bakes in a stationary worldview. New products look like “C” by design. Seasonal peaks look unimportant right before they hit. Segmentation freezes the past into buckets and projects those buckets into the future. The business, unfortunately, does not cooperate.
Most damaging, segmentation disconnects decisions from money. Assigning a class and then slapping a target service level onto that class is the opposite of pricing a choice by its expected return and risk. The class boundary becomes the decision rule; the economics become an afterthought. That is the wrong axis.
What to do instead
The remedy is not a “better segmentation.” It is to stop segmenting and start deciding at the finest useful granularity.
Concretely, this means three habits.
First, carry uncertainty explicitly. Single‑number forecasts force brittle rules; probability distributions let you see plausible futures and weigh them. Forecasts are a means to an end: they feed decisions rather than define them.
Second, rank all feasible micro‑decisions by expected dollar impact. Every extra unit, at every SKU and location, competes for the same capital. Build a score that combines the probability of sale with the relevant economic drivers—margin, cost of capital, stockout penalties, indirect basket effects, constraints. Then buy down the list until the next dollar no longer clears your threshold. This is mundane to compute and transformative in effect. It also eliminates the discontinuities that segmentation introduces, because priorities shift smoothly with the data rather than snapping at arbitrary borders.
Third, automate the mundane. If a method requires a weekly committee to apply it, it will not survive contact with complexity. Let software recompute priorities as often as the physical world warrants; reserve human attention for setting semantics and shaping the economics. Automation is not a quest for zero humans; it is how you make sure the humans are doing work that moves profit, not colors in a matrix.
This approach generalizes beyond inventory. In marketing, for example, you do not need “ten segments” to be practical; you need per‑customer decisions driven by behavior and tested for ROI. A decade ago we showed how to replace coarse segments with individual client analysis built on basket histories. The idea is the same: let the resolution of your decision match the resolution of your data and your economics.
Outgrowing the habit
Segmentation persists because it is human‑sized. It fits on a slide; it lets busy leaders feel they have tamed the mess. But modern computing has made the “human‑sized” constraint obsolete. When decisions are scored and ranked in money across the full option set, the need for buckets evaporates. The work moves to where it belongs: clarifying the economics, improving the constraints, and auditing a simple engine that does exactly what you told it to do.
If a practice adds more dashboards than dollars, retire it. The goal is not to win an argument about thresholds. The goal is to ship better decisions, every day, at scale.
Stop segmenting. Start deciding.
