What problem does it solve?
The reception desk of a large hospital fields the same questions all day: which doctor treats this complaint, where is the laboratory, is my report ready, when does the clinic open. In Tars's case study, Gama Hospital's IT director lists an information queue that had become an operational bottleneck in Saudi Arabia, with receptionists who lack the medical knowledge to route a symptom, so patients were sent from one doctor to another, and a call center of sixteen phone lines that was frequently engaged.
Language adds to it. Gama serves Arabic speaking patients and a large expatriate workforce, and the receptionist on shift may not speak the patient's language. Mappedin's case study on Princess Alexandra Hospital lists the same themes: complex layouts, staff under pressure and the need for multilingual help. A lobby assistant takes the repeated questions at the entrance at any hour, so staff keep their time for patients who need a person.
How does it work?
- The patient walks up and speaks. The assistant runs on a kiosk, a lobby tablet or an avatar screen. Voice input suits people who are standing, carrying things, elderly or uncomfortable typing in a second language. The agent detects the language and answers in it.
- It answers from the hospital's own data. Doctors mapped to specialties, department and floor locations, clinic timings and services form the knowledge it is allowed to use.
- It routes a symptom to a specialty. A described complaint such as back pain or headache maps to a specialty and the doctors available. Gama's agent points chest pain straight to Cardiology. A safer rule sends chest pain and other red flag symptoms to staff or the emergency department at once. Bangkok Hospital's registration kiosk uses an AI symptom checker to recommend a clinic.
- It gives directions. The answer names the floor and the nearest landmark, or gives verbal and visual directions on an indoor map, as the Princess Alexandra kiosk does with an indoor mapping SDK.
- It hands over. Anything urgent, unclear or outside the approved content goes to the front desk or a nurse.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- Kiosk and branch, Phone and voice, Digital human
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | about 800 | 1 | 1 vendor |
Value drivers: Lower cost to serve, Customer experience, Employee productivity, Inclusion and access.
Indicative value
A hospital with one to three lobby kiosks
USD 1200 to USD 54,000
Annual front desk time released per year
How this is calculated
Formula: kiosks * questionsPerKioskPerMonth * months * staffMinutesPerQuestion / 60 * staffCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Lobby kiosks in service kiosks, kiosks | 1 | 3 | Editorial assumption. Replace with your own number of entrances and devices. |
| Questions answered per kiosk per month questionsPerKioskPerMonth, questions per month | 400 | 1,000 | PAH about 800. Mappedin reports about 800 questions a month at one Princess Alexandra Hospital kiosk, a vendor report. |
| Months in service months, months | 12 | 12 | One full year. |
| Staff minutes avoided per question staffMinutesPerQuestion, minutes | 1 | 3 | Staff time avoided per question is an editorial assumption. Mappedin's case study mentions 3 minutes per conversation, so the range stays at or below it. |
| Fully loaded cost of one front desk hour staffCost, USD per hour | 15 | 30 | Editorial assumption, replace with your own payroll data. |
What it leaves out: Counts front desk minutes only. It leaves out platform and integration cost, content upkeep, the effect on phone line load, wrong routing avoided, accessibility gains and patient satisfaction. Whether the minutes turn into lower cost depends on staffing rosters.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Bangkok Hospital
Thailand · Healthcare · 2024
Bangkok Hospital built the Smart Registration system with Agnos Health, a medical AI startup, and runs it at the registration department of its headquarters hospital. Patients register at self service kiosks or on staff tablets. The kiosk checks identity by AI face recognition against an ID card or passport, runs an AI symptom checker that recommends a clinic, verifies insurance automatically and collects electronic consent. The hospital says the system aims to cut registration steps by up to 45%, which is a target, and publishes no measured result. This is a registration and clinic recommendation kiosk with an AI symptom checker, so it fits the page only in part: the description does not mention voice or open conversation.
No outcome disclosed.
Gama Hospital
Saudi Arabia · Healthcare · 2025
Gama Hospital, a multi specialty private hospital in Al Khobar, Saudi Arabia, launched the Gama AI Smart Guide in November 2025: a voice first AI agent on tablets in the lobby. Patients speak in Arabic, English, Hindi, Tamil or Malayalam, describe a symptom, and are pointed to a specialty, told when doctors are available and given the floor and nearest landmark.
The agent runs on the hospital's own mapping of doctors to specialties, departments, floors and clinic timings, and sends a severe symptom such as chest pain to Cardiology first. The vendor's case study reports that nearly half of conversations take place after 6pm and that patients ask more than three questions per visit on average. It publishes no containment or time saving figure.
No outcome disclosed.
Princess Alexandra Hospital
United Kingdom · Healthcare · 2025
Princess Alexandra Hospital in Harlow, England, uses a PRSONAS iHealthAssist avatar kiosk that answers visitors' questions and guides them to wards, rooms, services and amenities, with indoor maps from the Mappedin SDK. It gives navigation support around the clock in five languages, including British Sign Language. Mappedin's case study reports about 800 questions a month, of which about 35% are wayfinding and 65% general information. The figures come from a technology partner, not from the hospital.
- Interactions handled: about 800, per month, on average
"In one month, on average, 800 questions are asked and totaling 40 hours of interaction (estimate 3 minutes per conversation)."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Doctor, specialty, department and floor directory with clinic timings
- Symptom to specialty mapping approved by clinical leads, with urgent symptom rules
- Indoor floor plans and landmarks for directions
- Languages the patient population speaks, including local dialects
Systems to integrate
- Doctor scheduling or availability system, for live availability
- Indoor mapping or wayfinding platform
- Front desk or nurse call queue, for handover
- Analytics store for questions asked and unanswered
Complexity: Medium
The build needs a maintained directory of doctors, specialties, floors and clinic timings, a clinical owner who approves the symptom to specialty mapping and the urgent symptom rules, indoor maps for directions, and a kiosk or tablet in the lobby with a microphone that works in a noisy hall. Linking to live doctor availability adds an integration with the scheduling system.
- 1
Count what the desk is asked
Log a week of front desk and call center questions and group them. Start with the top four or five, which at Gama were symptom routing, doctor availability, medical report questions and wayfinding.
- 2
Get the symptom rules signed off
Clinical leads approve the symptom to specialty table and write the list of symptoms that always mean urgent staff contact. Put each rule in the test set.
- 3
Pilot at one entrance
Place one device where queues form, with a staff member close by. Review a sample of transcripts daily for the first weeks, including the hours after the desk quiets down.
- 4
Test languages with real speakers
Test each language and dialect with patients and interpreters, in the lobby's real noise, before adding it to the list.
- 5
Add live availability and maps last
Connect scheduling and indoor maps once the directory answers are right, and track which questions the agent could not answer.
Guardrails
- The assistant does not diagnose and says that it routes to a specialty only
- Urgent symptoms such as chest pain go to staff or emergency services first, with a clear message
- Answers only from the approved directory and policies, with a plain statement when it does not know
- AI disclosure on the screen and no storage of identifying health details in logs
KPIs to instrument
- Questions per day and per visitor, by hour and by language
- Share of sessions that end without staff contact, and share handed over
- Questions the agent could not answer, grouped by topic
- Wrong specialty referrals reported by patients or clinics
Human in the loop
A front desk or nurse is always reachable from the device and sees the transcript on handover. Clinical leads review a sample of routing decisions each month and the symptom table after any change to specialties or doctors.
Common failure modes
- A routed symptom hides a serious condition
- Keyword style routing can send a worrying complaint to a routine clinic. Keep a conservative urgent list, add a standing line that the guide does not diagnose, and review misses with clinicians.
- The directory goes stale
- Doctors move, clinics change hours and floors close. Give the directory an owner and a review date, and alert when answers cite entries older than the limit.
- Noise and accents break voice input
- A busy lobby degrades recognition for the languages that most need help. Test in the hall, keep a touch fallback, and track failed sessions by language.
- Patients treat it as clinical advice
- A confident answer reads as a medical opinion. Keep the wording to routing and information, and show how to reach a nurse.
What are the risks and rules?
EU AI Act
Depends on design
A guide that answers directory questions and gives directions is a limited risk system with transparency duties (Article 50). If it assesses symptoms in a way that influences care, it can count as medical device software, which the AI Act treats as high risk through Article 6(1) and Annex I when the device needs a third party conformity assessment. Whether symptom routing is medical device software depends on its intended purpose, so assess it under the EU MDR or UK medical device rules before launch.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be told when they interact with an AI system.
Controls to put in place
- AI disclosure on screen at the start of every session
- Clinical owner and review date for the symptom to specialty table and the directory
- Urgent symptom rules tested on every content or model change
- Masked, time limited logs with no storage of identifying health details
- Privacy impact assessment covering voice recordings in a public hall
Frequently asked questions
- Does a hospital lobby assistant diagnose patients?
No. Gama Hospital's agent takes a symptom and works out which specialty the patient should see. That is routing, not diagnosis; red flag symptoms such as chest pain should go to staff or the emergency department, under rules clinicians approve.
- What results have hospitals reported?
Mappedin reports about 800 questions a month at the Princess Alexandra Hospital kiosk, about 35% of them wayfinding. Tars reports that nearly half of the conversations at Gama Hospital happen after 6pm. Both are vendor reports, and neither gives a containment rate or time saved.
- Which languages can it support?
Gama's agent handles Arabic, English, Hindi, Tamil and Malayalam. Princess Alexandra's kiosk offers five languages including British Sign Language. Each language needs testing in the real lobby with native speakers.
- Is it high risk under the EU AI Act?
Directory answers and directions carry transparency duties. Symptom routing that affects care can make the product medical device software, which changes the risk class, so check the design with a regulatory adviser.
How to cite this page
Blits.ai AI Use Case Library, "AI assistant for hospital lobby questions, symptom routing and wayfinding", last verified 11 November 2026, https://www.blits.ai/ai-use-cases/hospital-lobby-and-wayfinding-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 11 November 2026: First published