Our own project · Meridian Hospital Group is a constructed operator; the instrument and its figures are real
Meridian Hospital Group · Automation
Smart records, cabinets, and The Hand
Automation in a hospital almost never removes the headcount the business case promised. What it actually returns is minutes at the bedside, traceability nobody had, and a measurement of the column this study says is never measured. It also converts frequent small problems into rare large ones, which is a good trade only if you kept the manual fallback.
Every automation proposal in healthcare arrives with a business case written in headcount, and almost none of them deliver it. That is not because the technology does not work. It is because the time it frees was already being borrowed from something else, so it returns as work finally done properly rather than as a smaller wage bill.
Which makes automation genuinely valuable and consistently mis-sold, and means the honest way to evaluate it is against the three things it actually changes: where the minutes go, what becomes visible, and how the thing fails.
Chapter 1 · Why the business case looks wrong
The saving is real and it is not headcount
Automate a task that a nurse was doing and you do not get a nurse back. You get a nurse who now has time to do the thing they were already supposed to be doing and could not: the observation taken properly, the conversation with the family, the check that catches the deterioration early. The ward gets measurably better and the wage bill does not move.
Finance then records the business case as unrealised, because the promised post never disappeared, and the organisation concludes that automation does not pay. That conclusion is an artefact of the measurement rather than a finding about the technology, and it is the misattributed cost from the entity model arriving from the opposite direction: a real benefit landing in a column nobody was watching.
So the case has to be written in what actually changes. Minutes returned to the bedside, expressed against the staffing arithmetic. Cases not cancelled. Stockouts avoided. Errors not made. Length of stay not extended by a decision that waited on information. Write it in headcount and it will fail on its own terms while succeeding on everybody else’s.
Chapter 2 · Four machines
What each of these actually is
Four things, and it is worth reading the last column of each. Every one of them fails in a specific way, and the failure mode is the part that determines whether the thing belongs in a hospital rather than in a warehouse.
-
Smart records
The instrument everything else reports into- What it actually does
- Structured clinical record, ordering, results and prescribing in one place, with every entry timestamped and every amendment visible as an amendment.
- What it really saves
- Not clerical time, which mostly moves rather than disappears. It saves the decisions that used to wait on a document, and it removes the repeat blood test, the repeated history and the eighth telling of the story.
- What it creates
- An audit trail, which is a safety control before it is an IT feature, and the raw material for measuring anything at all. Also alerts, which become noise if nobody prunes them.
- How it fails
- Silently and expensively. Downtime sends the ward back to paper, the record fragments for that period, and back entry afterwards is done by tired people. The soft spot in the record will matter enormously if anything from that day is ever disputed.
-
Drug vending and automated dispensing
Point of use supply- What it actually does
- Holds stock where it is used, releases against a prescription and an identity, and restocks on consumption rather than on a weekly guess.
- What it really saves
- Nursing minutes spent walking and searching, stockouts at three in the morning, and the expiry write-offs that come from stock nobody could see.
- What it creates
- A complete record of who took what, when, which is simultaneously a controlled drugs control, a workload measurement and a surveillance capability. All three are real and the third has to be named rather than discovered by the staff.
- How it fails
- As a locked box. A manual store degrades gracefully when the power goes or a fault is thrown; a cabinet does not, which is why the override path and the emergency key are part of the design rather than an afterthought.
-
Smart cabinets and tracked consumables
Knowing what you have, where- What it actually does
- Weight or tag sensed storage for implants, sets and high value consumables, recording issue and return against a patient and a procedure.
- What it really saves
- The search, the phone call, the duplicate order, and the case cancelled because the item that was on the system was not on the shelf.
- What it creates
- Cost per case that is actually traceable to a case rather than apportioned, which is the costing page getting the data it says it does not have. Also recall management that works, because you know which patient received which lot.
- How it fails
- Through workarounds. A cabinet that is slow or awkward at the moment of need gets propped open, and a propped cabinet is worse than a shelf because everybody believes the data.
-
The Hand
Store room assistant- What it actually does
- Automated picking and replenishment in the stores: the assistant retrieves, stages and returns, so that people request rather than search, and stock moves to the point of use on a schedule rather than on an errand.
- What it really saves
- The largest single unbooked cost on the people pages. Walking, searching and fetching are the work that porters, housekeepers, ward clerks and nurses absorb without it appearing anywhere, and this is the machine aimed directly at it.
- What it creates
- A measured picture of demand by location and hour, which is the input every levelling and just in time argument on this site has been asking for and which no hospital currently has.
- How it fails
- At the edges rather than the middle. It handles the regular and the catalogued; the urgent, the odd shaped and the one somebody needs right now still need a human, so the establishment cannot be cut to the automated case and then asked to cover the rest.
Chapter 3 · The mistake that makes automation worse than what it replaced
A machine that serves one person at a time is a queue
Here is the failure that gets designed in, costs almost nothing to avoid, and is missed constantly because it is a capital decision that is actually a queueing decision.
Medication is not evenly spread through the day. It clusters into rounds, and at the peak a ward generates far more requests than the average suggests. A dispensing cabinet serves one person at a time, which makes it a server with a service time, which makes a queue.
Run the same arithmetic this site used for beds and the answer is uncomfortable. At a realistic drug round peak a single cabinet on a ward runs at high utilisation and the nurses queue for it, which is the exact opposite of the benefit that was purchased. A second cabinet takes the wait to seconds.
The second cabinet is not a luxury and it is not gold plating. It is the difference between an automation that returns minutes to the bedside and one that takes them, and the decision between the two is made by somebody comparing unit prices who has not been shown this calculation.
derived: 28 medication events an hour at the peak, 2 minutes each, Erlang C
Chapter 4 · And the trade is only good if you kept the fallback
Automation converts frequent small failures into rare large ones
A manual store room fails constantly and gently. An item is in the wrong place, somebody walks further, a nurse borrows from the next ward. It is inefficient by design and it is extraordinarily robust, because every failure has a human workaround available immediately.
An automated store fails rarely and completely. Most days it is faster and more accurate than any person, and on the day it throws a fault or loses power it is a locked cabinet containing exactly the thing somebody needs now. The total amount of disruption may well be lower. Its distribution is entirely different, and hospitals are far more sensitive to the shape of a failure than to its average.
Which means the manual fallback is not a transitional arrangement to be retired once the system is trusted. It is a permanent part of the design: the override, the emergency key, the small buffer of stock held outside the machine, the ability to prescribe and administer on paper for a shift. Every one of those looks redundant on any day the system works, which is almost every day.
That is the spare again, in a fourth form. The organisation will be tempted to remove it once the automation has proved itself, and proving itself is precisely what makes it look removable. The reason to write the justification down at the point of purchase is that the person who deletes it later will not have been in the room.
The same argument in people is on staffing levels and the spare, and the power these machines depend on is on power and infrastructure.
Chapter 5 · Which is what this whole study has been asking for
The real prize is that they measure the unmeasured column
Every page on this site has run into the same wall. The patient’s waiting is not counted. The goodwill spent by staff is not counted. Where a delay was actually created is not counted. The three kinds of cost exist because the instruments point at one of them.
These four machines point at the other two, almost incidentally. A record with timestamps knows how long a decision waited on a result. A cabinet knows when the drug round actually started and how long it took. A tracked consumable knows what a case really used. The Hand knows how much fetching a layout generates, by hour and by location, which is the number the walking distance argument on the people pages had to assume.
That is a larger prize than the labour. An organisation that installs these and reads only the stock reports has bought expensive shelving. One that reads the timestamps has acquired, for the first time, an instrument pointed at the column that decides whether it is any good.
It has to be said plainly that the same data watches the staff, because it does, and pretending otherwise is how an organisation loses the trust that makes any of this work. The commitment that makes it legitimate is narrow and checkable: this data is used to find out where the process fails, not to rank individuals, and where it is ever used the other way somebody has broken a stated rule rather than interpreted a vague one.
Chapter 6 · Back to the decision
What it means for a hospital being built from nothing
A new build is the only time any of this is cheap. Retrofitting automated stores, cabinet positions, network and the power to run them into an occupied hospital costs several times what designing them in costs, and some of it is simply not possible afterwards because the space is not there.
So the decisions belong in the massing and the brief: where the stores sit and how the assistant moves between them, cabinet positions on every ward, the electrical and network provision to reach them, and the space for the manual fallback that will exist for sixty years. That is the same sentence as the infrastructure page, and the same sentence as the layout argument in the problem statement.
There is also a market specific point. In a setting with an unreliable grid, every one of these machines is a load that must be on the essential supply, and a locked cabinet during an outage is not an inconvenience but a clinical event. Automation raises the value of the power design, and the power design raises the cost of the automation. Those two decisions are one decision, and they are usually taken by different people in different years.