AI Agent for a Yoga Studio: Passes, Waitlists and Trial Follow-Up
A practical guide to running class passes, waitlists and trial follow-up through an AI agent that remembers each member's history within clear limits.

An AI agent for a yoga studio is a role-limited assistant that handles class passes, waitlist offers and trial follow-up while keeping a record of each member's history across visits and messages. It works through the booking system and approved channels, under rules the owner sets for refunds, exceptions and sensitive health details. The practical test is simple: when a member asks why they hold a waitlist spot or how many classes remain on their pass, the answer is correct without a front-desk person searching three different tools.
Key takeaways
- Start with one role, such as trial follow-up and waitlist offers, before giving an agent any question about pass billing.
- Keep member memory small and structured: trial status, pass balance, waitlist entries, class preferences and declared modification requests.
- Write approval boundaries for refunds, pass extensions, transfers and cancellation exceptions before the agent sends its first message.
- Choose one observable success criterion, such as whether every waitlist offer receives a reply within the same business day.
- Keep a named staff owner for exceptions and review the agent's action log on a fixed schedule.
Where studios lose members between the first class and the second
Most studios do not lose members in one dramatic moment. They lose them in small gaps: a trial visitor who gets no message after the first class, a pass that expires with two classes unused, a waitlist offer sent late in the evening and never answered. Each gap is minor on its own, but they accumulate because the information needed to act is spread across the booking system, the shared inbox, a messaging app and the memory of whoever is at the desk that day.
Front-desk staff are usually good at this work when they have time. The difficulty is that requests arrive in bursts, often during classes, and follow-up depends on remembering who said what and when. An agent is useful here because it can check the record at the moment a message arrives. Its value depends on that record being accurate and on the limits being written down before it runs.
Why member memory matters more than a scheduling chatbot
A chatbot that reads from a public timetable can say when the next morning class starts. It cannot tell a member that their introductory offer was already used, that they joined the waitlist for Thursday after a cancellation, or that they asked for a gentler sequence because of a knee. Those facts decide the right reply, and a reply without them is worse than no reply.
Memory for this kind of agent should be a short, structured record, not a pile of old transcripts. Useful fields include trial status, pass type and remaining count, class preferences, waitlist entries with timestamps, modification requests and open promises the studio has made. Each field should note where it came from and when it was last confirmed. The same design logic appears in AI for Yoga Studios: Memberships, Waitlists and Clear Limits, and a similar record structure is used for member histories in Superintelligent Agents for Golf Courses: Tee Times and Member Records.
Keep the health boundary firm. The agent can record that a member asked for modifications or mentioned a limitation. It should not judge whether a pose or class is safe for that person. That judgment belongs to the teacher or to the studio's own intake process.
A decision guide: what the agent answers and what staff decide
Define this split before any agent touches member messages. Write it as plain statements the team can check against real conversations:
- Agent answers directly: class times, studio policies exactly as written, remaining pass count, whether a waitlist spot is still open, booking confirmations.
- Agent answers and logs: trial follow-up messages, waitlist offers inside the stated response window, and suggestions to renew a pass that is close to expiring.
- Agent drafts, staff approves: refunds, pass extensions, transfers between members, exceptions to cancellation windows, price adjustments of any kind.
- Staff only: questions about injuries, pregnancy or medication, whether someone should attend a class, complaints, and disputes over charges.
The list is deliberately conservative. Widen the agent's permissions later, after the logs show it handles the routine cases well.
A step-by-step plan for the first role
A first engagement should cover one role, not the whole front desk. A sensible order of work looks like this:
- Write down the current rules for pass types, expiry, waitlist cutoffs and trial terms, in plain sentences that a new teacher could follow.
- Collect a sample of real recent conversations about trials and waitlists, including the awkward ones where a member was upset or confused.
- Define the memory fields, the source of each field, and the person responsible for keeping it correct.
- Set the approval boundaries from the decision guide and name the staff member who reviews drafts and exceptions.
- Choose one success criterion you can observe, for example whether every waitlist offer gets a reply within the same business day.
- Run the agent in draft mode, comparing its proposed replies with what staff would have sent.
- Turn on the narrow permissions only after the drafts meet the expected standard, and keep a log of every action the agent takes.
The draft period is where most of the learning happens. Mismatches usually reveal a missing rule, not a model failure, and fixing the rule is the cheaper change.
Illustrative scenario: a waitlist offer and a lapsing pass
Illustrative scenario: a member holds a ten-class pass with two classes left and about three weeks before it expires. A spot opens in the evening class she usually attends, and another member sits first on the waitlist. The agent checks the waitlist order against the studio's rules, sends the offer within the stated window, and records the response. When the first member replies that she cannot come that week, the agent logs the decline, offers the spot to the next person in line, and adds a note to her record that she prefers evening classes. It also notices that her pass is nearly expired and drafts a renewal message for the front desk to approve. The owner sees the full log the next morning and can check any step.
The scenario works because each action is small, rule-bound and visible. Nothing in it requires the agent to make a judgment about money or health.
How Be Human runs this as a managed service
Be Human starts with one defined role, the memory and tools that role needs, the approval boundaries and an observable success criterion, then operates it under human control. Managed operations include monitoring of the agent's actions, recurring evaluations against real member questions, maintenance when the booking system or messaging channel changes, periodic review of who and what the agent can access, and controlled changes when a rule or a recurring problem justifies one. The service is delivered remotely, in English or Portuguese.
Risks to plan for before go-live
Messages from members are untrusted input. A message can contain text designed to push the agent past a rule, such as a request to ignore the refund policy. For that reason, permissions must come from the system configuration, never from what the agent is told in conversation. The [OWASP guidance on prompt injection prevention](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) is a useful reference for the team building these safeguards. Authorization needs the same discipline: the agent should see only the members and actions its role requires, as set out in the [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html).
Keep an action log for every message and every change, so that any strange reply can be traced to a specific step. Decide how long conversation data is kept, who can read it, and how members can ask for it to be removed, and check those choices with counsel for your jurisdiction. Similar permission questions come up in other appointment businesses, such as AI for a Lash and Brow Studio: Fills, Rebooks and Patch Tests. Start with one role, keep the memory lean, and review it on a fixed schedule. If you want to find the first task worth handing to an agent in your studio, the free Be Human AI diagnostic is a practical place to begin.
Frequently asked questions
Can an AI agent manage class passes without replacing my booking system? Yes, in most studios the agent reads and writes through the booking system the studio already uses, within the limits it is given. It should not keep its own pass balances that drift away from the source of truth. Pass rules and expiry logic remain the studio's decision, and any change to them should be reviewed before the agent applies it.
Does the agent need to store everything members write? No. A short structured record with sources and dates is more reliable than full transcripts, and it limits how much personal data is kept. Health information should be recorded only as a request or limitation, with any clinical question routed to a qualified person.
Who decides when a conversation moves to staff? The studio decides, and the rule should be written before go-live. A reasonable starting point is that refunds, exceptions, injuries and complaints always go to a person, while routine status checks and waitlist offers stay with the agent.
How do I know whether the agent is working? Compare its replies with what staff would have sent on a sample of real conversations, and track the single success criterion you chose at the start. Review the action log on a fixed schedule, and treat any reply that promised something outside the rules as an incident to investigate and fix.
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.


