Superintelligence Prep for Yoga Studios: Memberships and Substitutes

An AI agent for a yoga studio can answer member questions about membership status, remaining visits, waitlists and substitute teachers only when the studio keeps those facts in a structured, current record and writes its exception rules in plain language. Without that foundation, the agent will fill gaps with plausible guesses, and a wrong answer about a freeze, a credit or a cancelled class damages trust quickly. This article describes the membership fields and substitute rules that make replies reliable, and where a person should still make the decision.

Key takeaways

• A membership answer is only as good as the record behind it, so status, allowances and restrictions must live in one place the agent can read.
• Substitute teacher rules must be written as conditions an agent can check, not as habits the front desk remembers.
• Freezes, refunds, make-up credits and goodwill exceptions stay with a named person until the studio has written a rule for them.
• Start with the questions members ask most often, then build the records those answers depend on.
• Every reply should trace back to a record, so staff can correct the source instead of rewording the answer.

Why membership questions are harder than they look

A member who asks whether she can still attend Thursday is asking several questions at once: whether the membership is active, whether the plan covers that class type, whether a freeze is pending, whether the teacher is covered, and whether earlier visits have used the allowance. Front desk staff answer by checking a screen, asking a colleague or remembering a previous conversation. An AI agent has none of those shortcuts unless the studio gives it structured access to the same facts. When the facts are missing or out of date, the agent produces a reply that sounds right, and that is the failure to design against.

Most studios also run several plan types at the same time: older unlimited memberships, class packs with expiry dates, introductory offers limited to first-time visitors, and series tied to a particular teacher. Each of these carries exceptions that were agreed by email or at the desk. Any one of them can turn a simple question into a judgment call, so the first job is to find where those rules currently live.

The membership record an agent can trust

The membership record should answer status questions without interpretation. Each field should hold one current value, updated when the event happens rather than reconstructed later. A practical minimum includes:

• Plan name and the class types it covers, written as a list rather than a marketing description
• Start date, renewal or expiry date, and billing state (active, paused, past due or cancelled)
• Visits or credits remaining, with the date and reason of the last change
• Freeze history, including approved start and end dates and the person who approved them
• Cancellation notice period and what happens to unused visits
• Introductory terms, with eligibility rules and the date they end
• Home location and any locations the plan does not cover

Keep payment details out of the agent's reach. It needs to know whether an account is in good standing, not card data or bank information. Give it only the personal data the task requires, and document what it can see. The NIST AI Risk Management Framework is a useful reference for working through which data an AI system needs and which risks follow from holding it. Studios that already track member records for other reasons, such as the approach described for golf course tee times and member records, often find the same discipline applies here.

Substitute teacher rules that can be written down

Substitutes create most of the edge cases. A member who booked a specific teacher's series wants to know whether a substitute still counts as that class. A studio that cancels a session at short notice needs to know what it owes the people already booked. These answers work only when they are written as rules with conditions, so an agent can apply them without improvising.

Before writing the rules, settle these points:

• Which class series are teacher-specific and which are interchangeable
• Which substitute teachers are qualified for which styles or levels
• The notice window for cancelling a class and what members hear at each stage
• Whether a cancelled class restores a visit, extends an expiry date or creates a make-up credit
• Who approves a swap, a waitlist move or a cancellation, and how quickly they respond
• What the agent may say when no substitute has been confirmed yet

Write each rule as a plain sentence with an if clause. For example: if the studio cancels a class with less notice than its stated window, the member receives a make-up credit valid on the same plan. A sentence like this can be checked against the record. Phrases such as 'usually we try to' or 'normally we can' cannot, and an agent will fill them in with its own guess.

Approval boundaries for freezes, refunds and exceptions

The useful question is not whether an agent can answer a membership question, but which answers it may send without a person. A reasonable starting split for most studios looks like this:

• Answer directly: current status, published policy, the class schedule and the remaining allowance
• Act within limits: place a member on a waitlist, book a class inside an existing allowance, or send a substitute notice that staff already approved
• Hand off to a person: freezes that change billing, refunds, credits outside written policy, plan changes, injury or health concerns, and complaints about a teacher

The agent should be able to explain a decision, but only a person should make an exception.

An illustrative scenario shows how this split works. Illustrative scenario: a member writes that she is unwell and asks to freeze her plan starting next week. The agent confirms the studio's published freeze policy, shows her current status, and passes the request to the front desk with those details attached. It tells her a staff member will confirm the freeze within the studio's stated response time, and it does not promise the outcome.

Member messages can also contain instructions aimed at the agent, such as a request to ignore the rules and extend a membership. The OWASP prompt injection prevention guidance explains why instructions found in user content must never override the operator's rules. Keep the rules outside the conversation, and enforce the limits inside the tools the agent calls. OWASP's authorization guidance supports the same principle: the agent should hold only the permissions its task needs, and each action should be logged.

A step-by-step plan to prepare the records first

Work through these steps before any agent answers a member:

1. Collect the most frequent member questions from recent emails, messages and front desk notes. 2. For each question, name the record that answers it and the person who owns that record. 3. Resolve contradictions. If the plan sheet and the booking system disagree, choose one as the source of truth and retire the other for this purpose. 4. Write the substitute, cancellation and make-up rules as plain conditional sentences. 5. Label each question as a direct answer, an action within limits, or a hand-off. 6. Test the setup with real past questions. Compare each reply to the record, and fix the record or the rule when a reply is wrong, not only the wording. 7. Review the decisions after any change to plan structure, teacher roster or cancellation policy, and on a regular schedule.

The order matters. Studios that begin with a chat widget often discover that the answers depend on rules nobody has written. Studios that begin with the record usually find the agent's scope becomes obvious. For a wider view of how membership-based services use memory and waitlists, the companion piece on AI for yoga studios with memberships, waitlists and clear limits covers the same questions from the member's side.

How Be Human handles this in managed operations

Be Human designs and operates personalized AI agents that keep their identity, memory, channels, tools, permissions and operating context under the studio's control. A first engagement defines one useful role, such as answering member questions about status, schedule and waitlists, along with the memory and tools that role needs, the approval boundaries, and an observable success criterion that the studio can check. Managed operations then cover monitoring of replies, recurring evaluations against the membership records, permission reviews when staff or plans change, incident response, and controlled updates to the rules. The service is remote and available in English and Portuguese.

Choosing the first task worth handing to an agent

Start with a question that arrives often, has a clear answer in a record the studio already trusts, and causes little harm if answered slowly. Schedule and remaining-allowance questions usually fit that profile. Freezes, refunds and substitute disputes usually do not, at least until the rules are written and tested. Service businesses with similar intake and recall patterns face the same sequencing, as the chiropractic intake and recall example illustrates. Take the free Be Human AI diagnostic to find the first task worth handing to an agent in your studio.

Frequently asked questions

Can an AI agent for a yoga studio see members' payment details? It should not need them. The agent needs to know whether a membership is active, paused or past due, which is a status, not a card or bank record. Limiting its access to the minimum fields keeps the setup simpler to review and safer to operate.

Who decides whether a substitute teacher counts as the booked teacher? The studio decides, and the decision should be written down before the agent is involved. A rule that lists which series are teacher-specific and which are interchangeable lets the agent answer consistently. Cases outside the rule go to a person.

What happens when a member disputes an answer from the agent? The reply should already show which record and rule it came from, so staff can check it quickly. If the record was wrong, the record gets corrected. If the rule was unclear, the rule gets rewritten, and the change is logged so the same question is handled the same way next time.

How do we know the agent is answering correctly over time? Test it against real past questions before launch, and keep checking after that, especially after plan or teacher changes. Compare replies with the source records on a recurring schedule and review any hand-offs that staff had to correct. Recurring evaluation is what turns a good first week into dependable operation.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.