The hospital systems nobody demos

Hospital software conversations are dominated by the clinical layer — EMR, order entry, clinical documentation. Reasonably so; it is what clinicians touch. But a hospital COO's week is rarely about the EMR. It is about a ward that ran out of a consumable while another wrote off the same item at expiry, a ventilator whose service contract lapsed, and a procurement process that negotiated a rate nobody applied.

We build that operational layer and we do not build the clinical one. Being explicit about the boundary is not modesty — EMR, clinical documentation, diagnostics and claims are regulated product categories with established vendors, and a firm claiming all of it is telling you something about how carefully it scopes work.

Consumables: central figure, local reality

Stock is usually recorded at the central store and consumed at the department. The recording of that consumption is where the system loses touch with reality — an issue slip filled at the end of a shift, or not at all, so the figure on screen and the quantity on the shelf diverge within weeks.

Once nurses stop trusting the figure, they start keeping a local buffer. That is entirely rational behaviour and it is fatal to the data: the store shows stock that is actually in eleven informal cupboards, and the reorder logic works from a number that means nothing.

Every ward with a private stash is telling you the central figure is not trusted. The stash is the symptom; the recording burden is the cause.

The fix is the same one that works everywhere: make recording faster than not recording. Scan at issue, at the point it leaves the store, on a phone or a fixed scanner at the counter. Par levels per department so the indent is suggested rather than composed. Batch and expiry captured at receipt so a near-expiry sweep is a query rather than a physical search.

Equipment: a register that drifted years ago

Biomedical equipment moves between departments — a monitor borrowed for a busy night and never returned, an infusion pump sent for repair and quietly replaced. Nothing records it, so the register's location field is a historical guess and the annual verification becomes a search rather than a check.

The consequences are more than administrative. You cannot compute cost of ownership, so replace-versus-repair is a judgement call. You cannot see which model of pump generates the most breakdowns, so purchasing repeats the mistake. And when a recall notice arrives for a specific batch of a specific device, finding them takes days.

The service contract that lapsed

This is the one that produces genuinely bad outcomes and it is almost always tracked in a spreadsheet. AMC renewals, warranty windows, calibration due dates and statutory inspection dates, held in a file that one person maintains and everyone else assumes is current.

Two failure modes. The expensive one: an AMC renewed late or twice, or a repair paid for on equipment still under warranty. The serious one: a critical device out of calibration during use, which is a patient safety issue and an inspection finding, not a procurement inconvenience.

Building this is not hard. Due dates as data, tickets raised ahead of time with an owner, vendors chased automatically, and downtime recorded against the asset so the renewal negotiation has evidence in it. It is unglamorous, cheap relative to almost anything clinical, and the return is immediate.

Procurement: the rate contract nobody applied

Rate contracts negotiated centrally and then not enforced at the point of purchase is the same failure as in construction, with the same cause: the person raising the requisition does not know the contract exists, and the system does not tell them.

Enforcement at requisition, with the contracted supplier and rate as the default and off-contract purchases requiring a recorded reason, converts a negotiated saving into a realised one. Supplier invoices matched against goods receipts and purchase orders, with batch and expiry captured at receipt, closes the loop and removes most of the manual checking from the accounts team.

Multi-site is where all of this compounds

One unit can survive these problems on institutional memory. Three or more cannot. Without a consolidated view, transfers do not happen — so one site writes off at expiry what another was short of that week. Equipment sits idle at one location while another hires in. And a recall means calling every site individually and hoping.

A shared inventory and asset picture across sites is usually the single change that pays for the whole programme, and it is worth being explicit that this is an operations project rather than an IT one. The system is the easy half; agreeing that a ward will scan at issue is the half that determines whether it works.

On data, and where it lives

Very little of this touches patient data, which is a genuine advantage — the risk profile of an inventory and asset programme is far lower than anything clinical, and it should be scoped that way rather than dragged into a clinical governance process it does not belong in.

Where hospital policy requires that data stays inside the building regardless, we deploy on-premise on your own hardware with the same access control, audit trail and pipelines we would build in a cloud. That is a deployment decision, not a compromise on how the system is engineered.