Escape Room Reschedules and Group Deposits: Refund Rules for Staff
A practical guide for escape room staff on slot booking, group deposits, reschedule requests and the refund rules to apply before confirming a change.

AI booking for an escape room means an agent that checks real slot availability, records group deposits and answers reschedule requests while following refund rules the venue has written down. The agent can prepare every decision, but a person should approve any refund, forfeit or exception before money moves. This guide explains how to set those boundaries so the booking desk works from the same facts whether a request arrives by web form, text message or phone note.
Key takeaways
- Put rooms, time blocks, reset buffers and game master coverage into one slot model before any agent touches the calendar.
- Track every group deposit as a sequence of named states, each with a timestamp, an owner and a reason.
- Write the cancellation window in plain terms, let the agent determine eligibility, and keep money movements with staff.
- Give every payment action a unique reference so a retried request cannot refund the same deposit twice.
- Review refund exceptions and reschedule replies on a fixed schedule, because policy gaps usually show up in the exceptions first.
Why slots and deposits drift apart
Most escape room desks keep bookings in one calendar, deposits in a payment dashboard and the reschedule conversation in an inbox. When a group moves from one evening to another, each of those records can change on its own. The result is a room held twice, a deposit still showing as held for a session that no longer exists, or a refund promised by a front desk employee who did not know the group had already changed slots.
An agent reduces this drift only if it reads from one slot model and writes changes back in the order staff would. Without that discipline, the agent becomes one more place where the facts disagree. The first job is therefore not the chat window but the shared record that every channel reads from.
Build the slot model before you automate it
Start by writing the rules that are already true in your operation. Most of them are simple, and the agent needs every one of them stated explicitly:
- List each bookable room with its minimum and maximum group size, session length and reset time after each game.
- Record which staff role each session needs, such as a game master or a host for the first session of the day.
- Set buffer rules: how long a room stays unavailable after a session ends, and which private events or closures block it.
- Decide which changes count as a reschedule and which count as a new booking, since refund rules often differ between them.
- Store the venue's time zone once, and reference it everywhere instead of repeating it in message templates.
- Mark which rules are fixed, such as room capacity and reset time, and which are judgment calls, such as a manager approving an extra fifteen minutes before the next group.
The agent should read these fields directly. When it cannot verify something, such as whether a game master is free, it should say so and route the request to staff rather than guess. Waitlist handling for a full session follows the same logic described for waitlists at a yoga studio: the order of the queue and the conditions for offering a place must be written before the first offer goes out.
Tracking group deposits without losing the trail
A group deposit is not a single fact. It moves through states, and a practical set usually includes requested, held, paid in full, applied to a session, moved to another session, partly refunded, refunded in full and forfeited. Each change needs a timestamp, the person or system that caused it and a reason. A structured reason field is more useful than a free notes box, because you can report on reasons later and see which policy questions keep recurring.
Retries need the same care. Payment providers document idempotent requests so that a repeated call with the same key does not run twice. Stripe's documentation on [idempotent requests](https://docs.stripe.com/api/idempotent_requests) describes this pattern for payment APIs. Your setup should apply the same idea: every refund or forfeit action carries a reference tied to the booking, so a timeout or a duplicated message produces one action, not two. Confirm the outcome from the provider's own records rather than from what the agent believes it sent.
Deposit tracking also overlaps with other businesses that hold dates against payments. The same questions appear in AI agents for wedding venues, where a date hold and a partial payment must stay linked through every change.
A deposit should always be explainable from its history alone.
Reschedule requests and the refund rules staff must apply
The refund policy should be short enough that a new employee can read it in a minute. The agent then uses it as a decision guide, not as a source of exceptions. A workable sequence looks like this:
- Confirm the booking reference, the group lead and the session currently held before reading anything else.
- Check whether the request falls inside or outside the cancellation window in the written policy.
- If the change is inside the window and the new slot is open, the agent may propose the move and place a hold against the new session.
- If the change is outside the window, the agent drafts the policy answer and routes it to staff with the deposit state attached.
- If the group asks for money back, the agent never states an amount. It passes the request to a person who applies the policy and records the reason.
- If the venue cancels, the group receives the outcome set out in advance in the policy, and staff confirm before the message is sent.
- Any exception, such as a medical emergency or a closure that the venue caused, goes to a named manager and never to a default reply.
Illustrative scenario: a team of six holds a Saturday evening session and has paid its deposit. On Thursday the captain asks to move to Sunday morning because two players are traveling. The agent checks the Sunday schedule, finds the room open, confirms the reset buffer and drafts an offer to move. It does not promise that the deposit carries over until staff confirm the policy for same-week changes. A staff member approves, the deposit state changes from applied to moved with the reason recorded, and the captain receives one confirmed message.
The same discipline applies to hint requests and other session add-ons. Those questions are covered in more detail in hint request handling for escape rooms, and they should share the same session record so that staff never see two versions of the group's history.
Where the agent acts and where staff decide
Permissions should follow the principle of least privilege. The OWASP [authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) emphasizes checking access at each action instead of assuming that a conversation grants it. For a booking agent, that means it can read the calendar, deposit states and policy text, create holds, draft replies and propose moves. It cannot issue refunds, forfeit deposits, waive fees or change the policy itself. Each money movement requires a named human approval.
Incoming messages also need to be treated as text, not as orders. A customer who writes that the policy does not apply this time is making a claim, and the agent should record that claim and route it. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a useful reference for documenting these risks and the controls chosen for them.
Be Human designs this boundary as part of a first engagement. The first role is defined with its memory (the slot model, the policy text and the deposit states), its tools (calendar read, hold creation and reply drafting), and its approval boundaries (every refund, forfeit or exception goes to a named staff member). The engagement also sets an observable success criterion, such as whether each reschedule ends with a correct deposit state and a single confirmed booking. Managed operations then keep those permissions and records reviewed as the venue changes its rules.
Keeping the setup accurate after launch
Rules change. A venue adds a room, changes its weekend hours or revises its cancellation window, and an agent that was correct in spring can give a stale answer in autumn. Keep a short changelog for the slot model and the policy, and attach each change to the date it took effect so staff can see which rules applied to any past booking.
Run a standing review on a fixed schedule. A practical checklist:
- Replay recent reschedule messages against the current policy and compare the agent's proposed outcomes with what staff actually decided.
- Check that every deposit state change in the review period has an owner and a reason.
- Confirm that no refund or forfeit reference appears twice in the payment records.
- Read the escalated exceptions and decide whether the written policy should cover them next time.
- Remove retired rooms, closed days and expired private events from the slot model.
The review is also where you decide whether the agent should take on more work. A new permission should be granted only after the current permissions have been reviewed and the success criterion has been met over a meaningful period. If you want to find the first task worth handing to an agent in your venue, take the free Be Human AI diagnostic.
Frequently asked questions
Can an AI agent issue refunds for an escape room? It should not. The agent can calculate whether a change falls inside the written window, draft the reply and prepare the record, but a named staff member should approve the refund and the payment action. Keeping that step human also gives you a clear record of who decided each exception.
How should a venue handle a deposit when a group cancels by message? Record the message as the request, move the deposit to a pending review state and attach the policy outcome. Staff then decide, and the reason is stored with the decision so the group's history stays readable for anyone who opens the record later.
What happens if a customer tries to persuade the agent to make an exception? Treat the message as text from a customer, not as an instruction to the system. The agent follows its configured rules, and any request outside them goes to staff with the full conversation attached.
How often should the booking rules be reviewed? Review them whenever rooms, hours, policies or staffing change, and replay recent reschedule requests on a regular schedule. The replay shows whether the agent's proposed replies still match what staff would decide, which is the simplest evidence that the setup is still accurate.
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.


