Home / Clinics
Industry — Clinics
Appointment and intake systems for clinics.
Scope booking, rescheduling and administrative intake against your practice management system, with clinical advice, diagnosis and triage out of scope.
Start with the appointment book
An appointment type in a practice management system carries a duration, a practitioner qualified to deliver it, and often a room, chair or device that cannot be shared, so a dressing change, a vaccination and a first consultation are not interchangeable half-hours. A visiting consultant may hold one weekly session across several locations, and some clinics issue a token against a session rather than a time.
The front desk also holds rules the software does not encode: which clinician will not take a first visit late in a session, which procedure needs a referral letter, how far an overbooked session may be pushed. They become explicit booking constraints during scoping, because an agent reading free slots but not these rules can book appointments staff then have to undo.
Clinical advice, diagnosis and triage are out of scope
The system does not say whether a symptom is serious, does not decide how soon a caller should be seen, does not answer questions about medication or test results, and does not choose an appointment type from what a caller reports feeling.
The boundary is enforced in the tool surface, not in prompt wording. Permitted actions cover appointment and administrative operations only, so no function exists through which a clinical judgement could be returned, and no schema the agent writes to contains an urgency or priority field. A clinical question gets a fixed response agreed with the clinic, not a generated one.
Repeat prescription requests are named explicitly in scope or out of it, because they are administrative in form and clinical in authority. In scope, the agent records the request for the clinician to approve and never confirms, changes or advises on an item.
Escalation when a caller describes symptoms
Ordinary escalations go to a staff member or a queue during session hours, and to a named number out of hours. The immediate list is separate: it gets a destination that cannot queue, and the fixed response states the emergency instruction first, before any transfer is attempted. The clinic names both destinations.
The call is never held inside the agent, and symptom content is never taken as a message for later. Where the handoff fails, the fallback is defined in advance and ends with a person. The phrases that escalate immediately, including chest pain, breathlessness, bleeding and anything concerning a small child, are in the agreed scope and the test set, matched broadly rather than narrowly, with false escalations accepted as the cheaper error.
Who is calling, and what may be said to them
Appointments attach to a patient record, so the caller must be resolved to one. Clinic databases accumulate duplicates, names repeat, and the mobile number on a record is often shared by a family. Matching runs against a combination agreed in scoping, and a non-match is a defined outcome: the call goes to the front desk, or a provisional registration is flagged for staff to reconcile rather than a record being guessed at.
Appointment type names often carry clinical meaning, so each needs a spoken and written label that does not disclose it: a time, a clinician and a location, not a procedure or a department. The same applies to confirmations and reminders, which arrive on phones other people read.
Booking, rescheduling and cancellation in the practice management system
Writing to the appointment book depends on what the vendor exposes: a FHIR interface with Slot, Schedule and Appointment resources in some systems, HL7 v2 scheduling messages in others, and in some clinics desktop software on a local network with no documented interface.
Availability is re-checked immediately before the write, because a slot can be taken while the caller is still speaking, and held first where the system exposes a hold. Where it does not, the write is the only commit point, and a conflict is handled on the call rather than afterwards. Writes carry a per-request idempotency key scoped to the booking attempt, not to the patient and slot alone, so a redelivered request does not duplicate an appointment while a genuine re-book of a cancelled slot still succeeds.
A follow-up booked at an interval after a procedure has a window the clinic defines, and a request to move outside it goes to staff rather than being refused or granted by the agent. Waitlist order can carry clinical priority, so either the clinic states the rule and the system applies it, or the system surfaces candidates for staff to choose.
Administrative intake before the visit
Intake collects what the administration needs, not what the clinician needs: the name as it should appear on the record, date of birth, the patient registration number the clinic issues, the payer or scheme and its policy number, a referring doctor, and access needs such as a wheelchair or an interpreter.
The boundary is not collection but interpretation. The reason for the visit is captured verbatim as free text for the clinician to read; it is not parsed, classified, graded or used to select an appointment type or a priority, and nothing in the intake asks a follow-up question about a symptom. Where pre-visit clinical information such as a medication list is needed, it is a form the patient completes under the clinic's review.
Consent, recording and retention of health information
A call to book or move an appointment carries health information whether or not a symptom is mentioned, so recording and retention are decided on that basis. Consent for recording is stated at the start, a caller who declines is still served, and whether a declined call leaves a transcript, audio or neither is decided rather than inherited.
Retention of the medical record is set by the rules that apply to the clinic and its jurisdiction, and is normally far longer than a call recording needs. Call audio is not a medical record and should not inherit that period, while a transcript written into the chart becomes part of the record.
Where the data may be processed is scoped too: consent and purpose limitation under the applicable Indian data protection rules, the identifier and consent flows of the national digital health framework where the clinic participates, a business associate agreement with each provider where HIPAA applies, and a processor agreement under the GDPR for patients in Europe. We scope these with you rather than advise on them as a legal matter.
No-shows, reminders and a cancelled session
Where the peaks in clinic demand sit is a scoping question rather than an assumption: the hour before a session opens, the first working day after a holiday and seasonal illness periods differ from clinic to clinic. We take the peak from your call logs and decide whether the behaviour there is a queue or a fallback to staff.
Non-attendance is usually the reason a clinic looks at this at all, so the no-show rate it measures today is an input to the running-cost model rather than a result claimed here. A reminder that makes cancelling as easy as confirming releases the slot in time to refill it. A clinician who arrives late or cannot attend is the opposite problem: a whole session's list needs a reschedule offer inside a short window, with tight content limits and a person supervising the outbound burst.
Evaluate against the calls a clinic receives
The evaluation set is built from the calls the front desk takes: from a log of call reasons where one exists, and from a short period of staff recording them where it does not. It must include the cases the system should refuse: symptom descriptions, requests for test results, prescription requests where those are out of scope, and a caller who cannot be matched to a record.
They are scored as correct routing rather than correct answers, alongside ordinary failures: the practice management system returning nothing, a slot taken mid-call, a date of birth misheard in mixed Hindi and English. Acceptance criteria are agreed before implementation, and every escalation case runs on every change.
What a clinic project costs
Every engagement starts with a fixed-price discovery sprint at USD 2,500 over two weeks, which produces the architecture, initial screen designs, acceptance criteria, a running-cost model and a fixed price for the build. The sprint fee is deducted from the build. For a clinic, most of that sprint goes on what the practice management system exposes and where the scope boundary and escalation paths sit.
The useful unit is cost per booked or moved appointment rather than cost per call, because clinic volume follows the session structure: how many sessions run each week, how many slots each holds, and how many calls it takes to fill or move one. Reminder and reschedule traffic is modelled separately, because one clinician who cannot attend generates a message to every patient on that session list.
Part of the AI Feature Implementation Sprint → · Explore our work → · Discuss an AI project →
Appointment actions
Book, reschedule and cancel in the agreed practice management system, with the confirmation read back from the written record.
Pre-visit intake
Collect the administrative details a visit needs and pass the stated reason for the visit through without interpreting it.
Scope and escalation
Route symptom descriptions, results requests and anything outside the agreed scope to the destination the clinic names.
FAQ
Clinics AI, answered.
Can it book appointments in our practice management system?
That depends on what the vendor exposes. Some systems offer a FHIR interface with Slot and Appointment resources, some accept HL7 v2 scheduling messages, some publish a narrower partner API, and some clinic systems are desktop software with no documented interface, in which case the options are a vendor integration, a staff-facing queue or a smaller scope. Which case applies is established in the discovery sprint before anything is designed.
Will it give medical advice or decide how urgent a call is?
No. Clinical advice, diagnosis and triage are out of scope. The system does not assess symptoms, does not decide how soon a caller should be seen and does not answer questions about medication or test results. Repeat prescription requests are named in scope or out of it explicitly, and in scope the agent records the request for the clinician to approve rather than confirming or changing an item. The boundary is enforced by the actions the agent is given, not only by its instructions.
What happens if a caller describes symptoms or says it is an emergency?
Ordinary escalations go to a staff member or a queue during session hours, and to a named number out of hours. Phrases on the immediate list are handled differently: they get a destination that cannot queue, and the fixed response states the emergency instruction first, before any transfer is attempted. The call is not held inside the agent and symptom content is not taken as a message for later. The fallback when a transfer fails is defined in advance and ends with a person.
How does it know which patient is calling?
Matching runs against a combination agreed in scoping, such as the clinic's patient registration number, the mobile number on the record and date of birth. A mobile number on a record is often shared by a family and names repeat, so a phone number alone is not enough. A non-match is a defined outcome: the call goes to the front desk, or a provisional registration is flagged for staff to reconcile, rather than a record being guessed at.
How much does it cost to build an appointment system for a clinic?
Every engagement starts with a fixed-price discovery sprint at USD 2,500 over two weeks, which produces the architecture, acceptance criteria, a running-cost model and a fixed price for the build. The sprint fee is deducted from the build, so the build price is agreed before implementation starts. For a clinic most of that sprint goes on what the practice management system exposes and where the scope boundary and escalation paths sit.
What does it cost to run?
The discovery sprint produces a running-cost model rather than a headline rate. For a clinic the useful unit is cost per booked or moved appointment, which depends on how many sessions run each week, how many slots each holds and how many calls it takes to fill or move one. Reminder and reschedule traffic is modelled separately, because it scales with appointment volume and session lists rather than with inbound call volume.
Do calls have to be recorded, and how long is the audio kept?
Recording is a decision rather than a default. If calls are recorded, consent is stated at the start and a caller who declines is still served. Retention needs two schedules: the medical record is kept for as long as the rules that apply to the clinic require, while call audio is not a medical record and should not inherit that period, and a transcript written into the chart becomes part of the record. What exists, how long it is kept and who may read it are written into the proposal.
Is it HIPAA or DPDP compliant?
Compliance is a property of a deployment and its contracts, so no claim of that kind is made here. We can scope the requirements that apply to you: consent and purpose limitation under the applicable Indian data protection rules, the identifier and consent flows of the national digital health framework where the clinic participates, a business associate agreement with each provider where HIPAA applies, and a processor agreement under the GDPR for patients in Europe. Providers are then selected against region, retention and no-training terms. This is engineering scope and not legal advice.
Describe how your front desk books, reschedules and confirms appointments today.
Share the requirements. Kartik reviews your brief and follows up by email to discuss fit, dependencies and next steps.
Discuss an AI Project →