Gym cancellations, freezes and plan changes with AI: the rules come first

Cancellations, freezes and plan changes are small, but they share three traits that make them good work for an AI agent and bad work for a busy owner:

• They follow rules. Notice periods, minimum terms, freeze limits and start dates are already in your membership terms. That is what makes them automatable.
• They have exceptions. An injury, a member who moved abroad mid-term, a charge they dispute. That is why a person stays in the loop.
• They touch money and goodwill. A wrong answer about a final payment costs more than the time it saved. That is why the rules have to be written down and tested, not improvised.

They also tend to arrive when nobody is at the desk: late at night, on Sundays, in the middle of a class. Answering them the same evening, correctly, is most of the value.

Most membership terms are written for a contract, not for someone answering a message. Turn them into a table: one row per request, what the rule must say, and who decides. The examples below are an imaginary box's rules; yours come from your own terms.

Then write down the things most terms leave out, because these are where a generic answer goes wrong:

• What counts as notice. Is a WhatsApp message enough, or does the member need to confirm by email or in the app?
• What the member gets back. A written confirmation with the dates, the last payment and what happens to their classes.
• The one alternative you offer. A freeze instead of a cancellation, or a smaller plan. Once, and only once.
• Who approves exceptions, and by when. So the member hears back within a day, not whenever the owner gets to it.
• What to say when the rule is unclear. "I'll check with the team" beats a confident wrong answer.

Writing this table is the hard part, and it is useful even if you never automate anything: your coaches and front desk will answer more consistently too. We explain how we write instructions like these in how to write instructions for AI agents.

With the table in place, an AI agent handles each request in the same five moves, whatever channel it came in on:

The hand-over is what makes members trust the agent, and what makes owners trust it too. Decide in advance what always goes to a person:

• Money: refunds, credits, disputed charges, anything that changes what a member pays outside the table.
• Health: medical freezes, injuries, anything a coach should hear about.
• Exceptions and feelings: a request the rules do not cover, or a member who is clearly unhappy.

Then decide how. The person you choose gets the conversation in the channel your team already uses, with a two-line summary and the documents attached; the member is told who will reply and roughly when; and once you decide, the agent closes the loop with the member and records the change.

Many messaging and membership platforms now include an AI assistant. They are a reasonable way to answer general questions (opening hours, what to bring to a first class). Membership admin is where owners tend to switch them off, and the reasons are usually the same, whichever product it is:

• It knows your FAQ, not your member. Without access to a member's plan and dates, it cannot tell whether this person is inside a minimum term or has already used both freezes this year.
• Prose rules get applied loosely. Paste in your terms as a paragraph and the assistant paraphrases them. A table, plus tests against real messages, is what makes the answer the same every time.
• There is no hand-over design. It either answers everything, including what it should not, or passes everything to a person, which saves nobody any time.
• It was never tested on your messages. General-purpose assistants are built to answer many businesses reasonably well, not to apply one gym's terms exactly.
• It cannot act. Unless it is connected to your membership software, it can only tell the member to email you, which is where they started.

None of this means a built-in assistant is useless. If yours lets you give it a rules table, a clear hand-over and access to member data, try that first: it may be enough. If it does not, the fix is usually the rules and the hand-over, not a cleverer model. We compare the options in a ready-made agent or a custom WhatsApp assistant and in build, buy or wait.

Automating cancellations does not mean making them easy to stop. Offering a member one alternative (a freeze, a smaller plan) is fair, and asking why they are leaving is useful. Making the exit hard is not, and the law is moving against it:

• California. Amendments to the state's automatic renewal law (AB 2863) tightened the rules for subscriptions taken out, changed or renewed from 1 July 2025: a member who signed up online must be able to cancel online, at will, through a clear link or button or a ready-made cancellation email, and any offer to stay must sit next to a "click to cancel" option (Fenwick; Barnes & Thornburg).
• United Kingdom. On 9 August 2026 the government said new "easy to exit" rules for subscriptions will come into force in January 2027 (GOV.UK). They come from the Digital Markets, Competition and Consumers Act 2024, and the detailed rules are still to be published (Baker McKenzie).

We are not lawyers, and the rules differ by country and state, so check what applies to your gym. What we build follows the spirit of both: the agent explains how to cancel, offers your alternative once, accepts a no, and confirms the cancellation in writing. The reasons members give go into a weekly list for you, which is worth more for retention than any script.

This is how we work with gyms: one job at a time, tested before it goes live, with the rules and the accounts in your name. The AI agent for gyms and CrossFit boxes page shows the whole member's year, from the trial class to the last message.