AI for Healthcare · Outpatient · Contact centre · Revenue cycle
Fill the clinic. Escalate what matters.
- Saudi Arabic, day and night
- Protocol-driven, never diagnostic
- Red flags go to a clinician
- PHI stays inside your network
04 · WHERE THE WORK LANDS
Bookings
→ Slot confirmed
written to scheduling
Waiting list
→ Gap refilled
in list order
Eligibility
→ Nphies checked
before arrival
Triage
→ Nurse connected
red flag · 8 s
Revenue cycle
→ Claim pack ready
documents attached
Follow-up
→ Patient called
concerns escalated
03 · HARNESS
VOHO HEALTHCARE AI HARNESS
ANSWER · IDENTIFY · ACT · ESCALATE
VOICE
Listen · speak · barge-in
TOOLS
HIS · slots · Nphies
MODELS
Arabic speech · LLM · docs
AGENTS
One per desk
IDENTIFICATION
Before anything is discussed
TRIAGE PROTOCOL
Yours, not the model’s
EVALUATIONS
Escalation accuracy first
MODEL ROUTING · CONTEXT · RAG · GUARDRAILS · PHI REDACTION · CLINICIAN ESCALATION · AUDIT TRAIL
02 · SYSTEMS OF RECORD
01 · HOW PATIENTS REACH YOU
Swipe the drawing sideways.
One AI layer. Every way a patient reaches you and every system behind it.
The Voho Healthcare AI Harness
A model with no scheduler is a very confident receptionist.
It cannot see the slot, the referral, the eligibility or the protocol, and it should never be the thing that decides whether chest pain can wait. The harness gives models the hospital’s systems, its identification rules and its triage protocol, and keeps clinicians in the only place that matters.
A model on its own
- Offers a slot that is not free
- Cannot see the referral or the approval
- Improvises clinical reassurance
- Has no way to reach a nurse
- Leaves nothing for the safety review
The same model, in the harness
- Books in the real scheduler, once the patient is identified
- Checks eligibility through Nphies before offering the slot
- Follows your written triage protocol, word for word
- Connects a clinician in seconds and hands over the context
- Flags every red-flag call for the clinical lead to review
Hospital systems
What you already run
- HIS / EMR
- Scheduling
- Nphies
- Insurance eligibility
- Claims
- LIS · results
- Pharmacy
- CRM
- Call centre
- SMS gateway
- Document archives
Voho Healthcare AI Harness
The runtime between your systems and the models
- Tool calling
- Model routing
- Clinical connectors
- Patient identification
- Triage protocols
- Memory
- Context
- RAG
- Escalation rules
- Permissions
- Human clinicians
- Evaluations
- Guardrails
- PHI redaction
- Audit trails
Model layer
Routed per task, never one model for everything
- Saudi Arabic speech
- Speech synthesis
- Foundation models / LLMs
- Document and ID models
- No-show and demand models
- Your own protocols
Agent layer
Scoped to one desk each
- Booking Agent
- Triage Agent
- Reminder Agent
- Eligibility Agent
- Results Agent
- Billing Agent
Where the work lands
Finished in the system that owns it
- Appointments booked
- Gaps refilled
- Eligibility checked
- Red flags escalated
- Paperwork read
- Follow-ups made
01
Answer.
Every call picked up in Saudi Arabic or English, evenings and weekends included, on the numbers your patients already have.
02
Identify.
The patient matched to the right file in the HIS with your own identification rules, before anything about them is discussed.
03
Act.
Booked, rescheduled, eligibility verified, referral logged, reminder sent: written into scheduling and the patient record, not a callback list.
04
Escalate.
Symptoms that match your red-flag protocol stop the booking and reach a clinician immediately. The agent never gives clinical advice.
01 · Safety first
The call where the agent refuses to book.
A patient asks for an appointment and mentions chest pain. The protocol your clinicians wrote stops the booking, the agent says only what the protocol allows, and a nurse is on the line in seconds with the context already in front of them.
Transcript
Identification
Nothing about a patient is discussed until identification passes.
Tool calls
waiting…
Policy
- Booking, moving, cancelling and explaining logistics are inside the delegated scope.
- Anything clinical stops here: no advice, no diagnosis, no medication. A red flag reaches a clinician in seconds.
Illustrative call · every red-flag call flagged for the clinical lead
02 · The empty slot
Cancelled at eight in the evening. Refilled by morning.
Nobody loses money on the appointments that happen. They lose it on the ones that cancel overnight, when there is no one to work the waiting list in order, check eligibility and write the change back into scheduling.
Today’s list
Waiting list · in order
Patient 4471
waiting 18 daysqueued
Patient 5512
waiting 16 daysqueued
Patient 6033
waiting 14 daysqueued
Patient 6190
waiting 11 daysqueued
Patient 6244
waiting 9 daysqueued
03 · The rest of the floor
Not one use case. The work a hospital runs on.
Bookings, triage, paperwork, dictation and the calls in between. Same harness, different desk. Play any of them.
1 of 6 · swipe for the rest
Press play to hear how this call goes.
Booking
—
Confirmation
—
After hours
—
Want this running in healthcare?
We build the first one with you, on your own systems, and it is usually working within a month.
Across the hospital
One harness. Every desk that answers a patient.
The telephony, the connectors, the identification rules and the protocol are built once. The next clinic is configuration.
Outpatient and bookings
The clinic that runs at 70% full
- Appointments booked, moved and cancelled directly in your scheduling system
- Gaps from cancellations refilled by calling the waiting list, in the right order
- Reminders and preparation instructions in the patient’s own language
- Insurance eligibility verified through Nphies before the patient arrives
- Referrals captured and routed to the right clinic with the paperwork attached
Patient services
The calls nobody has time for
- Results and report availability answered, without reading clinical detail aloud
- Visiting hours, preparation, directions and department questions
- Complaints logged, classified and escalated to patient relations
- Post-discharge follow-up calls, with anything concerning handed to a nurse
- Pharmacy refill requests routed to the right queue
Revenue cycle and back office
Where the money leaks
- Insurance cards, approvals and claim paperwork read and checked
- Pre-authorisation packs assembled with the clinical documents attached
- Rejections classified by reason and routed to the team that can fix them
- Policies, protocols and circulars answerable with the page cited
- Clinicians dictate a note and the record updates itself
Connectors
One AI layer. Every system behind the desk.
Nothing is replaced to get the booking line live. Read access first, write access only where you decide it belongs.
HIS / EMR
Oracle Health, Epic, Phoenix and others
Scheduling
Slots, clinics, waiting lists
Nphies
Eligibility, pre-authorisation, claims
LIS and results
Availability, not clinical interpretation
Pharmacy
Refill requests and queues
CRM
Patient relations and complaints
Telephony
Cisco, Avaya, SIP. Your numbers
Documents
Insurance cards, referrals, consent forms
Built for a hospital
What the medical director asks, answered in the product.
The boundaries below are configuration your own clinical and security teams set, and the escalation ones are tested before anybody goes live.
It does not practise medicine
No diagnosis, no advice, no medication guidance. It follows the triage protocol your clinicians wrote, and where the protocol says escalate, it escalates.
Red flags reach a person immediately
Chest pain, breathing difficulty, bleeding, pregnancy warning signs and the rest of your list stop the booking flow and go to a clinician or to emergency guidance on the spot.
PHI stays inside your network
On your own servers or in a Saudi region. Recordings, transcripts and patient data stay where your policy says, and the models run where the data is.
Identification before disclosure
Nothing about a patient is discussed until your identification rules pass, and what may be read out is configured per action, not left to the model.
Never trained on patient data
What patients say and send is used to serve that request, and is not used to train models, ours or anyone else’s.
Evaluated on your own calls
Before go-live the agents are scored against real past calls and against your protocol: what they would have booked, and every case they should have escalated.
From the library
What this would look like at companies you know.
Demonstrations, not customer work. We built five agents for each of these organisations from public knowledge of how they run; none of them asked, and none of them run this today. Open one to see this page’s argument applied to a hospital or insurer you recognise.
Ministry of National Guard Health Affairs
Healthcare
A national healthcare system where routine patient calls must be easy to complete and clinical exceptions must never be treated as routine.
Run the demos →Dr. Sulaiman Al Habib Medical Group
Healthcare
The largest private healthcare provider in the Kingdom, where most patient contact is about an appointment, an approval or a bill.
Run the demos →Mouwasat Medical Services
Healthcare
The Eastern Province’s largest private hospital group, serving an industrial workforce whose healthcare arrives through their employer.
Run the demos →Al Hammadi Holding
Healthcare
A hospital operator whose revenue depends on insurance claims being coded correctly the first time.
Run the demos →Dallah Healthcare
Healthcare
A hospital group where the call that decides the day is a patient trying to reach a clinic that is fully booked for six weeks.
Run the demos →Fakeeh Care Group
Healthcare
A hospital group where emergency department flow decides whether the rest of the hospital works that day.
Run the demos →Saudi German Health
Healthcare
A multi city hospital network where the same patient can be treated in three cities and their record follows them only if somebody carries it.
Run the demos →King Fahd Hospital of the University
Healthcare
A university teaching hospital’s outpatient clinics, where a cancelled session sends people home who travelled across a province to attend.
Run the demos →How a project runs
Start with the booking line. Keep the harness.
The first project proves itself on numbers you already report: answered calls, no-show rate, gaps refilled, and escalation accuracy.
01
Take the booking line
One clinic or one call type, where the queue is longest and the measure is obvious: answered calls, no-show rate, gaps refilled.
02
Connect inside your network
Scheduling and eligibility first, read access to the HIS, your telephony, reviewed by information security and by the medical director.
03
Replay real calls, protocol first
Scored against calls that already happened, with every red-flag case checked explicitly. Escalation accuracy is the number that decides go-live.
04
Go live out of hours, then widen
Evenings and weekends first, where the alternative is a voicemail. The next clinic reuses the same harness.
What clinical and operations leaders ask first.
If your question is not here, bring it to a technical session with our engineers.
01Does it give medical advice?
No. It books, reschedules, reminds, verifies eligibility and answers logistics. Where a caller describes symptoms, it follows the triage protocol your clinicians wrote, which mostly means getting a person on the line rather than giving advice.
02What happens if someone calls with chest pain?
The booking flow stops. The call is escalated to a clinician or to emergency guidance immediately, according to your protocol, and the whole exchange is recorded and flagged. Escalation accuracy on cases like this is tested before go-live and monitored after it.
03Where does patient data live?
On your own servers or in a Saudi region, as your policy requires. Recordings and transcripts stay inside that boundary, only named staff can read them, and nothing is used to train models.
04Can it check insurance before the visit?
Yes, through Nphies and your existing eligibility integrations, so a patient is not turned away at reception and your team is not chasing approvals on the day.
05Will it read results out to patients?
It confirms whether results are ready and routes the patient to the right clinician or channel. Clinical content is not read out by an agent unless your medical leadership explicitly configures a case where that is appropriate.
06Our patients speak many dialects and switch languages.
They speak Najdi, Hijazi, Egyptian, Levantine and others, often mixed with English. The agent handles the switching and reads dates, times and numbers back the way a person would.
07How do you handle no-shows?
Reminders go out on your schedule, and when someone cancels, the waiting list is called in the order your clinic decides until the slot is filled. Everything lands in the scheduling system, not in a spreadsheet.