Superintelligence Prep for Concierge Medicine: Membership Admin Only
Which membership, scheduling and billing-admin tasks a concierge practice can hand to AI agents, and which medical decisions clinicians keep.

Superintelligence prep for concierge medicine means defining AI agents that handle the membership, scheduling and billing-admin work around a practice, while physicians and clinical staff keep every medical decision. The right first step is to define one bounded admin role, the memory and tools it needs, the actions it may never take alone, and a measurable success test. This article lays out that scope for a practice leader who has to decide what to hand off first.
Key takeaways
- Start with membership administration: enrollment questions, renewal reminders, scheduling requests and billing explanations.
- Write clinical decisions into the role definition as off-limits, and route any clinical content to a named clinician at once.
- Give the agent only the records and approved documents it needs, each with a named owner.
- Keep money movements as drafts or staff-approved actions until the logs show the role behaves as specified.
- Judge the first role by one observable criterion, reviewed on a fixed schedule.
Which concierge admin work can an agent take on
Concierge practices run on a small number of recurring administrative tasks that repeat for every member and take staff attention out of proportion to their risk. Most of them depend on approved language and clear records rather than clinical judgment, which makes them good candidates for a bounded agent role. The work usually falls into a few groups:
- Answering membership questions from approved plan language, such as what the membership includes, how visits are counted and how a family member is added.
- Preparing renewal reminders and lapsed-membership follow-ups as drafts for the front desk to send.
- Handling scheduling requests inside approved templates, including confirmations, reschedules into open blocks and waitlist offers.
- Collecting administrative details, such as changed contact information or a new payment method, and routing them to the right staff queue.
- Explaining a statement in plain language once staff have confirmed the charge is correct.
Repeat-visit workflows in other health settings follow a similar pattern. AI for Acupuncture Clinics: Intake, Packages and Follow-Ups shows how package and follow-up work can be scoped in the same way.
Where the clinical boundary sits
The boundary should be written before the agent sees a single message. Any request that asks for interpretation, advice or a change to the medical record belongs to a clinician, and the role definition should say so in plain terms. In practice, the following stay with licensed staff:
- Symptom triage, including whether a message needs same-day attention.
- Advice about medications, doses, test results or procedures.
- Changes to diagnoses, problem lists or care plans.
- Waivers, exceptions or refunds that depend on a clinical reason.
- Discharge or termination of a member for clinical reasons.
Messages from members are untrusted input. They can mix clinical content with administrative requests, and they can contain text that tries to override the agent's instructions. The OWASP guidance on prompt injection prevention treats this as a design problem, not only a wording problem, so the boundary belongs in the workflow itself. Clinical keywords and patterns should trigger a handoff rather than depending on the model to judge correctly every time. For a clinical setting with its own scheduling limits, AI for Behavioral Health Clinics: Intake and Scheduling Limits shows how those limits can be written down before any automation starts.
A step-by-step plan for the first membership workflow
Choose one workflow and build it end to end before adding a second. The sequence below lets a practice stop at any step without leaving a half-connected system behind.
- Name the workflow and its owner, for example renewals for the annual membership, owned by the practice manager.
- Write the allowed outcomes in plain language, then write the forbidden ones beside them.
- List every system the agent may read and the single action it may take in each, such as creating a draft or opening a ticket.
- Define the escalation rule: which message patterns go to which named staff member, and how quickly that person is expected to respond.
- Test the role on past messages with identifying details removed, or on messages written by staff who play the member, before any live use.
- Review the logs on a fixed schedule, correct the instructions or source documents, and widen scope only after the success criterion holds across that review.
What the agent must remember and where it gets its answers
Memory and retrieval are different, and the difference matters in a medical practice. Memory is what the agent keeps from earlier sessions, such as a member's stated preference for morning calls. Retrieval is the lookup of approved documents at the moment of an answer, the pattern that retrieval-augmented generation reference architectures describe. For membership administration, the answer source should be the current membership handbook and the approved fee schedule, not whatever the model absorbed during training or whatever a staff member once typed into a chat.
Keep memory small. Store only the fields the workflow needs, such as membership status, renewal date and preferred contact channel, and do not copy clinical notes into the agent's memory at all. Assign a named owner and a date of last review to each source, so that when the handbook changes, the agent's answers change with it. The NIST AI Risk Management Framework is a useful reference for this kind of documentation of data, purpose and ownership, which early setups often leave out.
Billing, permissions and audit trails
Give the agent the narrowest permissions the workflow needs. Read access to membership status is usually reasonable. Writing to a billing system is not, unless the action is limited, reversible and approved in advance. OWASP's authorization guidance stresses checking permissions on every action rather than trusting a request because it arrived through a familiar channel, and that principle applies directly to an agent that receives messages from many people.
Several practical rules follow from this:
- Charges, refunds and fee waivers require a named staff approval recorded in the log.
- Payment event notifications are verified before the agent acts on them, as payment providers document for their webhook deliveries.
- Repeated requests must not produce repeated charges, so billing actions should carry an idempotency key, the pattern providers such as Stripe describe for safe retries.
- Every action records the instruction version, the source document, the message received, the draft or action produced, and the person who approved it.
The log is what makes review possible. Without it, a practice can see that a member was unhappy but cannot tell whether the agent, the template or the staff handoff caused the problem.
Illustrative scenario: a lapsed renewal and a fee question
Illustrative scenario: A member's renewal date passes with no payment on file. The agent places the approved reminder draft in the front desk queue instead of sending it, because the membership terms require a staff call before any lapse notice. Later the member replies to ask whether a visit last month counted toward the annual fee. The agent answers from the handbook section on visit counting and states that it cannot confirm an individual charge. When the same member asks for a refund, the agent opens a ticket with the message and billing record attached, tells the member that a staff member will reply, and promises nothing. If the reply had mentioned chest pain or difficulty breathing, the agent would have shown the practice's approved urgent-contact message and alerted a clinician immediately, without answering the billing question at all.
How Be Human runs membership admin under clinician control
Be Human designs, implements and operates this kind of role as managed infrastructure. A first engagement defines the single membership role, the memory and tools it needs, the approval boundaries and an observable success criterion. Managed operations then cover monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements, so the role stays within the limits the practice set as its systems and policies change. Service is remote, in English and Portuguese.
Take the free Be Human AI diagnostic to find the first task worth handing to an agent.
Frequently asked questions
Can an AI agent answer membership questions for a concierge practice? Yes, when the answers come from approved plan language and the agent is limited to questions the handbook covers. Anything that asks for a clinical judgment, a refund or an exception should be routed to a named staff member. The practice should test the answers on realistic messages before members see them.
Who is responsible for clinical decisions when an agent is part of the workflow? Licensed clinicians remain responsible for every medical decision, and the agent has no authority to triage, advise or change a care plan. The role definition should list those decisions as excluded and send any clinical content to a clinician at once. Responsibility for the workflow itself stays with a named practice owner.
Should the agent be allowed to issue refunds or change charges? In most practices, not on its own. The safer pattern is that the agent prepares a summary and a draft, and a staff member approves the money movement, with the approval recorded in the log. Narrow automation of reversible, pre-approved actions can be considered later, once the logs show the role behaves as specified.
How do you know the first agent role is working? Define one observable criterion before launch, such as the share of renewal questions answered correctly from approved language without staff rework, and review it on a fixed schedule. Check the logs for escalations that should have happened but did not, as well as those that happened unnecessarily. If the criterion does not hold, correct the instructions or source documents before widening the scope.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


