Get started

Rescheduling and cancellations

Changing an appointment sounds simple until the caller has several bookings, the calendar has different rules for each service, or the requested time is no longer available. Staff must identify the right appointment, explain the allowed options, update the calendar, and make sure the old slot is released. A cancellation may also need a reason, a follow-up task, or a human decision about fees.

An AI phone agent can handle the routine path without pretending every request is routine. It can ask for the details needed to locate a booking, use the calendar actions you expose, confirm the change, and route exceptions to a person. That makes it a narrower workflow than general appointment scheduling: the agent is modifying an existing commitment, so identity checks and change policies matter more.

A concrete call walkthrough

Consider a caller who needs to move a Tuesday appointment to later in the week.

Agent: I can help with an existing appointment. What name is the booking under, and what is the scheduled date and time?

Caller: Jordan Lee, Tuesday at 10 a.m.

Agent: Thank you. Before I make a change, please confirm the phone number attached to the booking.

After the caller provides the required matching information, the agent uses the connected calendar action to find the appointment. If more than one result matches, it asks a disambiguating question rather than guessing.

Agent: I found the Tuesday appointment. Would you like to reschedule it or cancel it?

Caller: Reschedule. Do you have anything Thursday afternoon?

The agent requests only the availability that the business has chosen to expose. It offers eligible times, the caller selects one, and the agent submits the change. It then reads back the new date and time and confirms that the previous slot is no longer booked. If the update fails, the agent does not announce success; it explains that the change is not confirmed and offers a human handoff or callback.

A cancellation follows the same controlled pattern. The agent confirms the appointment, states the approved policy language, asks for a cancellation reason only if the business needs one, and submits the cancellation. It should never decide whether to waive a fee or override a cutoff unless the connected system provides an explicit, approved action for that case.

How to set it up on ThunderPhone

Start with a prompt that defines the workflow as a sequence, not a vague instruction to “manage appointments.” Include:

  • the fields required to locate a booking;
  • the identity checks required before a change;
  • which appointment types may be changed by phone;
  • the rescheduling window, eligible locations, and approved policy language;
  • the exact confirmation details to repeat back; and
  • the conditions that require a transfer or staff follow-up.

Connect Cal.com or Google Calendar and expose only the actions the agent needs for availability, rescheduling, and cancellation. Keep policy decisions outside the calendar tool. If customer or case context matters, use HubSpot or Salesforce for the associated record. Google Sheets can support a deliberately simple change log, while REST or webhooks can pass structured outcomes into an internal system. Webhook endpoints can have separate secrets and event subscriptions, and ThunderPhone signs requests with HMAC-SHA256 so the receiving service can verify them.

Use an eligible production number for the live workflow. A demo U.S. number is inbound-only and intended for testing, not production. If staff should take exception calls, configure cold transfer or attended warm transfer. Warm transfer requires an eligible production number that can place an outbound SIP leg; the agent can privately brief the recipient and return to the caller if the recipient does not accept.

Before launch, build scenarios for duplicate names, missing bookings, timezone confusion, fully booked days, calendar write failures, late cancellations, and callers asking staff to waive a policy. ThunderPhone supports browser mic tests, AI-caller simulations, reusable scenarios, graded call logs, regression suites, and live-traffic A/B experiments. Simulations are billable real calls, with the charge shown before the run.

What to measure

Track the workflow outcome, not just whether the call was answered:

  • the share of eligible requests completed without staff intervention;
  • reschedules confirmed versus attempted;
  • cancellations recorded with the required reason or disposition;
  • failed calendar actions and duplicate-booking incidents;
  • transfers caused by policy exceptions or identity mismatches;
  • time from call start to a confirmed calendar state; and
  • later no-shows or repeated change requests by workflow path.

Compare those results with appointment reminder calls and no-show reduction rather than treating every schedule problem as the same one.

Honest limits

The agent is only as reliable as the booking data and actions it can access. It should not infer which of two similar appointments the caller means, invent an opening, or treat a failed write as a completed change. Staff should handle disputed fees, unusual accommodations, angry callers, requests involving sensitive judgment, and any case where identity cannot be established under the business's policy. Businesses such as dental offices may also need their own notices, privacy controls, and escalation rules around appointment details.

The safe boundary is clear: let the agent execute approved calendar operations, and send exceptions to a person with the context already collected. That is the practical boundary between automated appointment setting and staff-owned policy judgment.

Configure a ThunderPhone agent for rescheduling and cancellations