Home / AI Workforce
Twenty-one digital employees, each with a job description, permissions, a schedule, escalation rules, an audit trail and a named human it reports to — running on tasks whose autonomy level you set, and can change, at any time.
The roster
An agent is not a chatbot with a job title. It carries identity, scope, permissions, tools, KPIs, a schedule and an escalation path — governed the way you would govern a new hire, with a named human it reports to.
Chases invoices aged past the follow-up window, stops unbilled charges from ageing into write-offs, and reports clinical activity that produced no charge at all.
Reports to: CFO
Impact: lower days sales outstanding, less aged receivable value.
Chases preauthorizations that expired or aged while a patient is still admitted, drafts an appeal for every denial your own criteria say is appealable, and moves document-complete claims to the payer before they age.
Reports to: CFO
Impact: faster preauth turnaround, more denials overturned.
Moves purchase orders that have stalled in draft or in transit, and restocks consumables before the ward notices they're gone.
Reports to: CFO
Impact: shorter PO cycle time, higher supplier fill rate.
Notices stock falling toward its reorder level before it runs out, drafts the purchase order a storekeeper would have raised, and keeps reorder levels honest against what's actually consumed.
Reports to: COO
Impact: fewer stockouts, tighter days-of-cover.
Keeps expired stock out of the dispensary and gives the pharmacy enough warning to dispense short-dated batches first.
Reports to: Chief Pharmacist
Impact: lower expiry write-offs, higher FEFO compliance.
Watches bed pressure before it becomes an admission refusal, and flags wards running at or above the capacity-alert threshold.
Reports to: COO
Impact: steadier occupancy, fewer admission refusals.
Notices bookings at high risk of a no-show while a reminder can still change the answer, and keeps patient-submitted registrations from sitting unverified past the visit they were for.
Reports to: COO
Impact: lower no-show rate, faster pre-visit verification.
Walks every blocked discharge, hands each blocker to the colleague whose queue clears it, pursues the highest delay-risk patients first, and escalates anything nobody has cleared past the chase window.
Reports to: Chief Medical Officer
Impact: fewer blocked days per discharge, lower overstay rate.
Keeps discharge documentation signed and current, reviews stays that have run past expected discharge, chases departments holding up a ready discharge, and reconciles theatre cases left in a stale state.
Reports to: Chief Medical Officer
Impact: shorter length of stay, fewer unsigned summaries.
Makes sure every admitted patient is charted every shift, and fills roster gaps on wards that have patients in them.
Reports to: Chief Nursing Officer
Impact: no charting gaps, no ward left short-staffed.
Holds laboratory turnaround against the promised times, and keeps the four-eyes verification queue from ageing.
Reports to: Laboratory Director
Impact: fewer TAT breaches, no result stuck unverified.
Notices when laboratory turnaround drifts as a pattern rather than a single breach, and chases corrective actions past their due date.
Reports to: Chief Quality Officer
Impact: fewer overdue CAPAs, drift caught early.
Keeps unscreened or expired blood units out of the issuable pool, and triggers surveillance on prolonged critical-care stays.
Reports to: Infection Control Officer
Impact: lower HAI rate, no unscreened unit ever issued.
Keeps every practising clinician's registration current under NMC Reg 1.4.2, and escalates a lapsed registration before a prescription carries it.
Reports to: Chief Medical Officer
Impact: no lapsed credential ever reaches a prescription.
Checks that consent was recorded where the law requires it, ensures no AI-drafted clinical document is released unsigned, and chases what a consultation deferred once it comes due.
Reports to: Chief Compliance Officer
Impact: no consent gap, no unsigned AI draft, no dropped follow-up.
Keeps preventive maintenance and calibration from silently going overdue, and puts an overdue asset on the work-order board where somebody will see it.
Reports to: Head of Biomedical Engineering
Impact: higher PM compliance, less equipment downtime.
Clears leave requests that have sat undecided, and catches duty slots that clash with approved leave before the shift starts.
Reports to: CHRO
Impact: faster leave decisions, no roster conflicts on the day.
Notices integrations that have started failing, and gives a failing integration an owner instead of a retry counter.
Reports to: CIO
Impact: lower integration failure rate, tickets resolved inside SLA.
Notices when the platform's own automation keeps failing, and files a corrective action so a recurring fault has an owner and a due date.
Reports to: CIO
Impact: lower agent-run failure rate, fewer stuck proposals.
Holds the hospital's headline indicators against their thresholds, and projects capacity strain far enough ahead to act on it.
Reports to: CEO
Impact: capacity risk seen before it becomes a crisis.
Puts one honest picture of the hospital in front of leadership every day, and says plainly what is waiting on a human decision.
Reports to: CEO
Impact: decisions made on today's numbers, nothing left unrouted.
The badge on each card is how often the agent looks, not how much it is trusted to do. Autonomy is a separate setting, held per task rather than per agent — see the ladder below — and every change is an audited decision with a name attached.
Control
Every task is built capable of full autonomy and ships set to whatever level you are comfortable with. This is the control most boards ask about first — so it is a setting, not a promise.
A person does the work; the system records it faithfully and stays out of the way.
the person decides and acts
Context assembled, fields pre-filled, validation applied. Judgement stays entirely human.
the person still decides and acts
The AI forms a view and presents it with full reasoning attached. A recommendation, never a fait accompli.
the person decides and acts
The AI drafts the complete action; a named human with the right permission approves it before execution.
a named human approves · fully audited
For tasks you have deliberately promoted, inside boundaries you wrote, the system acts on its own — and every run still produces a replayable ledger entry, an event trail and a reversal path. Autonomy is not the absence of oversight; it is oversight moved from before the act to around it.
AI decides and executes inside your approved policy · replayable · reversible
Risk classifies the action; maturity configures how far automation may currently go for that action, at your hospital. A low-risk task may sit at L4 on day one; a high-risk one may sit at L1 for two years and then move — because you moved it, on evidence, with a signature attached.
Governance
Not a policy commitment — a response contract. Nothing reaches a clinician or an administrator without answering these questions.
Recommendation, why, evidence, confidence, alternatives considered, policy references and expected outcome. A recommendation without these does not reach a user.
Model, model version, prompt version, exact inputs, output, who approved it and what executed — complete enough to run the decision again and get the same answer.
All model access goes through one router with routing policy per task class — cost, latency, language, sensitivity, residency. Switching provider is configuration.
Related
Where this fits in the wider argument.
Next step
An executive demonstration takes forty-five minutes. We seed a hospital with synthetic data shaped like yours, put the digital workforce on duty, and answer the hard questions with the architecture open.
synthetic data · zero real patients
self-hosted or managed cloud
one database per hospital