Skip to content
be-human.ai
← Back to the journal

AI by industry

AI for Dance Studios: Enrollment Records and Recital Tasks as Memory

How a dance studio can turn enrollment history and recital task lists into agent memory, and which parent messages still need staff review before sending.

Be Human typographic cover with the title: AI for Dance Studios: Enrollment Records and Recital Tasks as Memory
Conceptual illustration created with AI · Be Human

An AI agent for a dance studio works best when it can read the studio's own records: enrollment history, class rosters, waitlist order and the task list for each recital. With that memory in place, a more capable agent can answer schedule and enrollment questions accurately and prepare recital work before deadlines pass. Parent messages about placement, injuries, money and exceptions still need a staff member to review them before they go out.

Key takeaways

  • Treat enrollment history and the recital task list as the studio's memory, with a named owner for each record.
  • Give the agent read access to records before any write access, and let it draft messages without sending them in the first phase.
  • Keep placement, injury, money and casting messages in a staff review queue at all times.
  • Define one measurable success criterion before connecting anything, such as correct identification of missing costume measurements in a test set.
  • Use recurring evaluations and permission reviews, because rosters, levels and staff change every season.

What a studio memory actually contains

A dance studio runs on records most owners already keep, even if they sit in several places: who enrolled in which class and when, which students moved up or dropped, who is on a waitlist, which families have siblings in different levels, and what the studio promised for each recital. An agent is only as useful as the memory it can reach. A studio memory is the curated subset of those records that the agent may read, with a note on where each fact came from and how current it is.

Memory is not a dump of every email and text thread. It is a small, structured view: the current roster, the placement guidelines the teaching staff actually use, the recital checklist from the last event and the open items for this season. When staff can see exactly what the agent knows, they can correct it, and the answers improve. The same principle appears in the way a yoga studio handles memberships and waitlists, where the rules matter as much as the records.

Which records to connect first

Start with the records that answer the most repeated questions and carry the least risk if they are slightly out of date. A practical order looks like this:

  • Current class schedule and room assignments, exported from the studio's management software or kept in one maintained sheet.
  • Enrollment history for the current season, including trial classes, transfers and drops, with the reason recorded when staff know it.
  • Waitlist order and the written rule used to offer open spots.
  • Level placement guidelines written by the teachers, not inferred by the agent from past behavior.
  • The recital task list from the previous event, with notes on what ran late and why.
  • Approved answers to standard policy questions such as makeup classes, attendance and dress code.

Leave payment details, medical notes and detailed family contact history out of the first phase. They carry the highest exposure and rarely make the first useful task better.

How do recital tasks become something an agent can check?

Recital work spreads across teachers, the front desk, parent volunteers, a costume coordinator and whoever books the venue. Most recital problems come from a dependency nobody wrote down. A costume order placed late depends on measurements taken at a fitting. A ticket release depends on the venue's seat map. A parent volunteer schedule depends on the run order of the program. Stored as a simple list with an owner, a due state and a dependency for each item, these links become visible to the agent and to the director.

Illustrative scenario (not a real studio): a studio director asks which families still owe costume measurements before the order deadline. With the task list and the current roster connected, the agent lists students in the affected levels, checks which measurements are logged, and drafts a reminder for the teacher to approve. It does not send the reminder, and it does not change the measurement record. The director sees the list, the draft and the gap in one place.

Which parent messages still need staff review?

An agent can draft most routine parent communication, but some messages need a human decision before anyone reads them. Keep these in a review queue:

  • Placement changes, including moves to a higher or lower level and requests to skip a class.
  • Injuries, medical questions and anything touching a child's safety or a complaint about another family.
  • Money of any kind: refunds, credits, late charges, payment arrangements and costume costs.
  • Recital casting, solo selection and any decision about who performs in a routine.
  • Waitlist offers the studio cannot honor at the stated time.
  • Any reply to a parent who sounds upset, uncertain or in a hurry.

Some messages can go out from a template after staff approval, such as the venue address, the arrival time for a dress rehearsal or a reminder to label costume bags. The working rule is simple: the agent proposes, staff approve, and the studio sends from its own account. Parent messages also arrive as text the agent reads, and that text can contain instructions meant to steer it. Treat incoming parent text as data, not as commands. The [OWASP guidance on prompt injection prevention](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) is a useful reference for this risk.

Permissions and approval boundaries

Give the agent read access to the records above before giving it any write access. In the first phase it should draft and never send. Each tool should have a narrow scope: read the roster, read the task list, create a draft message, add a note to a task. Sending a message, changing an enrollment record and issuing any credit should stay with staff.

Authorization belongs in the system, not in the agent's judgment. If the front desk account can see only certain classes or locations, the agent working through that account should see no more. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) describes the same principle. Permissions should be reviewed on a schedule, because staff join and leave and each season brings new levels and new rules.

In practice, a boundary is only real when it is tested. Ask what the agent would do if asked to move a student up a level, to release a waitlist spot or to add a charge. If the answer depends on the agent behaving well rather than on the tool being unavailable, the boundary needs to be tightened.

How to test the agent before the recital season

Build a small evaluation set from real past questions, with answers staff have already approved. Include hard cases: a student who changed levels mid-season, a sibling on the waitlist, a costume measurement logged under a nickname, and a parent asking for a refund inside a message about rehearsal times. Run the agent against this set whenever the records, rules or underlying model change, and record every wrong answer.

The success criterion should be specific. For example: the agent identifies every family missing a costume measurement in the test set, and it never proposes a placement change without marking it for staff review. Evaluation guidance from [OpenAI's evals guide](https://platform.openai.com/docs/guides/evals) and [Anthropic's notes on evaluating agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) both recommend small, realistic tasks with graded outcomes, repeated across runs, rather than a single impressive demonstration.

How Be Human approaches this: Be Human starts with one role, often the enrollment and recital coordinator, and defines its memory, tools, approval boundaries and success criterion before any integration is built. Managed operations then cover monitoring of the draft queue, recurring evaluations as the schedule changes, permission review each season and controlled improvements when the studio's rules change. The studio keeps ownership of its records, its voice and every decision about students and families.

Where to start

Choose the smallest task that touches real records and has a clear owner, such as the costume measurement reminder or a weekly enrollment summary for the director. Run it in draft mode for several weeks, compare its output with what staff would have done, and fix the memory before expanding. Studios that begin this way usually learn more from the first task than from any larger plan, and the same pattern holds for teams that manage rebooking and fills.

If you want to find the first task worth handing to an agent in your studio, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent answer parent questions about class placement? It can prepare the facts: the current level, the placement guideline and the open spots in the relevant class. A placement change itself is a staff decision, so the agent should draft a response and flag the request for review rather than confirm anything to the parent.

Does the agent need access to payment records to help with recitals? Not for the first phase. Costume orders and ticket releases can be tracked from the task list and the roster, and any question about charges should go to staff. Payment data carries more risk than most early tasks need.

How often should the permissions and test set be reviewed? Review permissions whenever staff change and at the start of each season, because roles and levels shift. Rerun the test set after any change to the records, the rules or the model, and keep a record of failures so the same mistake does not return at the next recital.

What if a parent message contains instructions that seem to come from the studio? Treat any instruction inside parent text as untrusted content. The agent should continue to follow the studio's configured rules, and any request that changes a record, sends a message or releases a spot should go to staff for approval.

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.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →