Lesson 03 — The morning batch

Warehouse movements concentrate at 06–08. That’s the data-entry shift, not the consumption window.

Plot a hospital’s warehouse movement events by hour-of-day and roughly 70% post between 06:00 and 08:00. After 11:00 there are practically no movements until the next morning.

The first reading is “the warehouse only operates in the morning”. That reading is wrong. Wards consume materials all day — gauze, syringes, drugs, dressings flow out as patients are treated. What the data shows is not when consumption happens, but when consumption is posted to the system.

The actual workflow

1. Throughout the day, ward staff pull materials from satellite stocks. Paper requisition slips or PDA scans accumulate.
2. Overnight, warehouse and pharmacy logistics staff collect the requisitions, reconcile against satellite stock balances, prepare the batch.
3. At morning shift start (06:00–08:00), the batch is entered into SAP / IS-H all at once. Thousands of rows post in a two-hour window because that’s the data-entry shift, not the consumption window.

The compromise behind this is operational and well-understood: warehouse staff cannot scan every consumption event in real time — wards are too busy, the systems too slow, the patients too distributed. Overnight reconciliation and morning-shift batch posting is what most Swiss hospitals settled on.

What changed in the analysis

The signal that benefited most from this Wisdom was signal_billing_before_usage. The signal flags cases where billing posts before the usage event — a real workflow risk that masks revenue leakage. But with the morning-batch artefact, a billing at noon and a usage at 07:00 the next day looks like a 19-hour lag, which looks like a workflow gap, which is in fact the posting cycle doing what posting cycles do.

Three concrete changes followed:

1. Day-granularity aggregation, not hour. The right unit for warehouse movement timing in this domain is the day, not the hour. The Lab visualisation switched accordingly.
2. Outlier detection on time-of-day was retired. It was returning artefacts, not signals.
3. Late-evening movements got their own attention. The few movements that do post after 18:00 are typically emergency restocks or pharmacy compounding. A minority pattern, but worth flagging on its own terms — not as outlier-against-baseline.

What to watch for

• Hospitals with real-time barcode scanning at consumption point show movements distributed across the day, not concentrated in the morning. That’s a different operational maturity, not a defect. The signal still fires honestly there; the artefact is absent.
• A break in the morning concentration (a Tuesday with movements distributed evenly) often signals a system outage that forced manual entry — a data-quality event worth investigating in its own right.

Why this is in a Lessons journal

Because temporal patterns in operational data are usually posting patterns, not event patterns. The distinction is invisible from the data shape alone. Knowing the operational workflow is what converts the data shape from anomaly into context. The Wisdom doesn’t change the signal’s SQL; it changes how anyone reads the output.

Sources

• Wisdom: smebit_warehouse_morning_batch.yaml
• Provider: a hospital warehouse manager with operational responsibility for logistics
• Anchor signal: signal_billing_before_usage
• Date Wisdom captured: 2026-05-08