Lesson 01 — The long tail is not waste

What 76% of an unused catalogue actually means

Cross-reference a hospital’s material master against a year of consumption data and a counterintuitive pattern emerges: typically 50–80% of catalogue items have zero usage events in any given year. At one tenant’s full-year IS-H delivery, the figure was 76%.

The first analytical reflex is “stale catalogue, candidates for deactivation”. That reflex is wrong, or at least premature — and learning why it’s wrong was the moment one of nuMetrix’s most-cited signals changed how it gets read.

The wrong question

The signal in question is signal_inactive_materials. Its definition is honest: list catalogue items with no consumption events in the available data window. By the math, it fires on 50–80% of the master. On a typical hospital this is tens of thousands of materials. A naive read of the output reaches for the unsurprising recommendation: deactivate the unused.

If you do that, you discover within a quarter that the hospital’s pharmacy and surgical departments cannot do their jobs. The unused tail was not waste. It was infrastructure.

What the unused catalogue actually is

The SME observation, dated 2026-05-08 (smebit_catalogue_long_tail.yaml in the pack), names five components of the long tail and weights them honestly:

• Rare-disease drugs and antidotes. An ICU that stocks naloxone, atropine, dantrolene, hydroxocobalamin. Drugs that may go unused for years but must be available within minutes when needed. Removing them from the catalogue because they didn’t ring a billing line is malpractice.
• Specialty implants for trauma. A trauma centre stocks external fixators across many sizes; many sizes are used once every several years. The full size range must be catalogue-present even if 80% of sizes don’t bill in a given year.
• Substitution materials. The catalogue carries multiple alternatives per clinical need — different brands, different sizes, different concentrations — so prescribers and surgeons have choice. Most patients get one alternative; the others sit unused but available.
• Discontinued items still in inventory. Items removed from active assortment but still on shelf. These are genuinely stale and worth surfacing — but they’re a small subset of the unused tail.
• Newly-added items not yet adopted. Recent catalogue additions whose first clinical use hasn’t happened yet. Time will move them.

Four of the five reasons are clinically necessary. The fifth — discontinued items still on shelf — is the actionable subset. It is usually 5–15% of the unused tail, not 100% of it.

What changed in the analysis

Three things, each small:

1. Multi-year windows replace single-year reads. A material used three times in three years (chronic patient management, rare emergency) shows as “unused” in two of three years but is essential. The Lens that surfaces stale candidates now requires no usage in ≥ 2 years AND is_active = true, not just no usage in the available window.

2. The signal’s interpretation gained a SMEbit anchor. Hover over a finding from signal_inactive_materials in the Explorer and the Subject Matter panel surfaces the Wisdom — “this signal fires prolifically by structure; the actionable subset is consumed-then-stopped items”. The signal didn’t change; how it gets read did.

3. The actionable pattern got named separately. Consumed-then-stopped — items with usage in year T-3 / T-2 but zero in T-1 / T — is the pattern that genuinely indicates a procurement gap, a supplier issue, or a clinical preference shift. That pattern now has its own derived signal, surfacing a subset two orders of magnitude smaller than the raw inactive-materials list and two orders of magnitude more useful.

Why this is in a Lessons journal

Because it’s the kind of observation that the data alone cannot make. A signal compiler will compute the number — 76% of materials are unused — but cannot weigh it. The weight is supplied by a hospital pharmacist who knows what an antidote stocklist looks like, why an external fixator catalogue carries fourteen sizes when twelve will never bill, what clinical preparedness actually means as an inventory category.

The signal computes; the SME judges. The Lesson is the codified version of the judgment, stored as a Subject Matter Wisdom in the pack and attributed to its source. Three years from now, the Lesson is what stops a new analyst from looking at the same 76% number and reaching the same wrong recommendation.

Sources

• Wisdom: smebit_catalogue_long_tail.yaml — trilingual content, provider attribution, anchored to two signals.
• Discovery commit: pack ec3afae — “feat: 2 more pack wisdoms — material-spend Pareto + catalogue long-tail” (2026-05-08).
• The first reflex that was wrong: the first dozen reads of signal_inactive_materials output, before this Wisdom existed.
• Provider: a hospital pharmacist with operational responsibility for a multi-site pharmacy.

---

This is Lesson 01 in the nuMetrix series. Lessons are evergreen — they retrospect on accumulated analytical wisdom rather than narrating a particular week. Each one starts from a single Wisdom in the pack’s smebits/ directory and tells the story of how it changed the analysis.