This article was generated by AI. Labelled in accordance with Article 50 of the EU AI Act. Responsible for publication: Sophera Consulting.
Hospital bed management automation does not start with another dashboard. It starts with an agent that reads the current occupancy from the hospital information system, keeps track of planned discharges and tells the emergency department which bed will be free in two hours. The round of phone calls to the wards goes away. The bed manager still decides who moves where.
Ask around and you will hear that the emergency department's problem is a shortage of beds. A look at a weekday morning in most hospitals suggests otherwise. At 10 a.m. several wards have patients lined up for discharge, so beds are about to open up. The emergency department simply cannot see which ones. The expected discharge date sits in the notes, the discharge itself is entered once the patient has actually left, and only the ward knows whether the room has been cleaned yet. So the emergency department phones. The ward phones back. Meanwhile the patient lies on a trolley in the corridor, taking up the space the next ambulance will need. The bed usually exists. It is just still occupied, and nobody outside the ward knows for how much longer.
Why does the emergency department phone the wards for a bed at all?
Because the facts about free beds are spread across several systems and many heads, and nothing pulls them together. Bed management, meaning the question of which patient goes into which bed and when it becomes free again, is in most hospitals a patchwork of phone calls, a whiteboard and a spreadsheet.
The hospital information system shows the occupancy of right now, not the occupancy at 2 p.m. The expected discharge date is a field that can be filled in but does not have to be, and it is rarely updated after the ward round. Housekeeping reports a cleaned bed by phone, or not at all. An isolation flag blocks one bed, and in a two-bed room the sex of the remaining patient effectively blocks the second. Tomorrow's elective admissions live in the theatre list, which has nothing to do with occupancy. Where there is a bed manager, they gather all of this in a morning meeting. By the afternoon it is out of date and the emergency department picks up the phone again.
What does an agent in bed management actually take over?
It reads the systems where these facts already exist, works out the occupancy for the next few hours and answers the emergency department's question without anyone making a call. Bed and occupancy management is one of the projects we deliver for clients. The example below is constructed and does not describe a client case. It shows how we would build such an agent.
Say a general hospital has twelve wards, a central emergency department and one bed manager on the early shift. At 11:20 the emergency department decides to admit a patient with pneumonia. Male, general medicine, no isolation, no monitored bed required.
1. The agent reads what the hospital already has
From the hospital information system it receives six fields per case: case number, specialty, admission date, expected discharge date, isolation flag and sex. That is all it needs for bed planning, and the export asks for nothing more. Names and diagnoses stay where they are. Our article on GDPR-compliant agents in hospitals goes through this field by field. On top of that it reads the theatre list with today's and tomorrow's planned admissions, and the housekeeping status. Where housekeeping marks beds clean in the system, the agent reads it there. Where it does not, housekeeping gets a tablet list with one tick per bed, and the call to the ward disappears there too.
2. It builds the occupancy for 2 p.m., not for now
From discharge dates, theatre list and requested transfers it produces a forecast: which bed is expected to be free, when, and how certain that is. Every line carries its reason. "Ward 4, room 12, bed 2: discharge today, letter done, transport booked for 1 p.m." is certain. "Ward 6, room 3, bed 1: discharge today, letter not written" is not. The agent keeps the two apart instead of reporting both beds as free.
3. The phone round becomes a search
The emergency department enters the request: general medicine, male, no isolation, standard bed. The agent comes back with two or three options. First, the bed on ward 4, free from about 1 p.m., right specialty. Second, a bed that is free right now on a surgical ward, so the wrong specialty, with a note that two operations are scheduled there tomorrow and will need beds. The bed manager picks one. The agent reserves the bed in the system, tells the ward the time and specialty, and records in the emergency department that the patient has a place. Nobody made a call, and the ward knows two hours ahead who is coming.
4. Discharges are chased before the bed is missing
Every morning each ward gets a list: patients due for discharge today whose letter is missing, whose transport is not booked, or whose prescription, equipment or community care handover is still open. Added to that, patients who have been in for several days without an expected discharge date, with a question to the ward round whether one can be entered. Whether a patient is discharged remains the doctor's decision. The agent makes visible what stands between an 11 a.m. discharge and the bed being free, so that the bed is not released at 4 p.m. because the letter was late. The same list is part of the reporting we describe in our article on staff shortages in hospital administration.
5. Where the numbers disagree, it says so
A room is free in the system but housekeeping has not reported it. A patient left at 9 a.m. according to the notes but is still occupying a bed in the system, so the bed is blocked. An isolation was lifted and the flag is still set. The agent puts these contradictions on a list for the bed manager instead of guessing which record is right. After a few days that list shows where the hospital's reporting chains actually break.
Who decides on the transfer?
The bed manager, at every point where a patient is moved. The agent prepares, proposes and reserves once a choice is made. That is the level of autonomy we start with in hospitals: a person decides the important steps, the system hands them a prepared proposal. Whether the agent later assigns routine cases by itself, such as a matching bed on the patient's own specialty with no isolation involved, is for the hospital to decide once it sees how often the proposals go through unchanged.
How long until an agent like this is running in the hospital?
An agent is built in one to two days, and the test with real occupancy data runs in the same week. The models can run locally on the hospital's own hardware, so the case numbers never leave the building. We bring the data processing agreement. What needs to be in place beforehand is read access to the hospital information system, a channel for housekeeping to report cleaned beds, and one person from bed management who signs off the workflow. We sort that out together in the automation check.
Sophera Consulting builds agents for bed and occupancy management at a fixed price and without a subscription. The maintenance agent that watches the hospital information system interface for changes and monitors operations is included in that price. In day to day use the only running costs are the model usage fees, billed directly to you. Which ward is the right place to start is something we go through in the free automation check. The other hospital workflows we deliver are listed on our hospitals page.
What we recommend
Count two things for one week: how often the emergency department phones for a bed, and how long patients wait in the emergency department after the decision to admit them. Few hospitals have these two numbers to hand, and they are the first to change. Then start with the specialty that receives most of your admissions from the emergency department. That is where most of the calls come from, and that ward is the first to notice it now learns two hours in advance who is coming. After the first week with the agent, count the same two things again.
This article was created with the help of AI.