Almost every appointment automation project starts with the same objective: reduce the no-show rate. The question we ask in the first meeting usually moves the project one step backwards.
How are you calculating that rate?
The answer is frequently not singular even inside one organisation. Operations quotes one number, finance another, branch managers a third. The gap is not measurement error. It is a difference in definition.
Fix the definition first
A no-show calculation has at least four distinct cases to separate:
- The customer did not attend and gave no notice
- The appointment was cancelled shortly before the slot
- The appointment was cancelled early enough for the slot to be refilled
- The organisation cancelled, because of staffing, equipment or capacity
Put these in one bucket and the resulting number means nothing. An early cancellation is actually a good outcome; the slot was recovered. Reporting an organisation-side cancellation as customer behaviour is self-deception.
The denominator needs settling too. Is it total appointments created, confirmed appointments, or appointments still active? When automation goes live, appointment creation volume usually rises because the channel opens around the clock. The denominator grows, the ratio falls, and the fall gets read as improvement.
Writing these definitions down at the start of the project is the only antidote to the "did it really improve" argument three months later.
Reminders work, but find out how much with your own data
The effect of reminders on attendance is one of the better-studied questions in healthcare. The Cochrane systematic review of mobile phone messaging reminders found evidence that they improve attendance (Gurol-Urganci et al., 2013).
The content of the message has been measured too. In two randomised controlled trials by Hallsworth and colleagues, stating the cost of a missed appointment to the health system in the reminder reduced missed appointments (PubMed).
These findings give direction; they do not give you your number. Effect size varies with sector, how far in advance appointments are booked, customer profile and reminder channel. So we recommend holding back a control group when the reminder flow goes live: a portion of appointments receives no reminder, and the difference is measured over the first quarter. Skip that and all you are left with is "the rate went down", with no way to separate automation from seasonality.
Timing is more than one decision
Whether a single reminder or two works better depends on what the reminder is for.
A reminder sent too early, a week ahead, is forgotten. One sent too late, two hours ahead, reminds the person but leaves the organisation no room to rebook.
The structure that works in practice has two stages: one early enough to allow cancellation or rescheduling, one close to the appointment day. The first exists to recover the slot; the second exists to secure attendance. Because they do different jobs, their content should not be identical.
The gain is in cancellation, not attendance
This is the part projects most often get backwards. Reminder flows are usually designed to persuade the person to attend, and cancellation is deliberately made difficult.
The result runs the other way. The person who is not going to attend still does not attend; the organisation simply finds out at the appointment time, and the slot is wasted.
A customer who tells you they cannot make it is valuable. The earlier they tell you, the more valuable, because that slot can be filled from a waiting list or new demand. So one of the first options a reminder flow offers should be some version of "this time does not work, shall we change it".
The advantage of a voice assistant here is doing three jobs in one call: confirming, recording the cancellation, and proposing a new time in the same conversation. With a message-based reminder the customer cancels and the process stops there; rescheduling requires a separate contact.
For the same reason, the metric to watch is not only the no-show rate but the refill rate for released slots.
Two technical traps in calendar integration
Double booking. The assistant offers a slot while talking to the customer, the customer thinks about it, and in the meantime the same slot fills from another channel. At the end of the call the record cannot be created, or worse, it creates a clash.
The fix is to hold the proposed slot briefly for the duration of the call. Too short and the customer cannot decide; too long and capacity is locked unnecessarily at peak hours. A hold window of a few minutes covers most scenarios in automated flows. When the hold expires the slot must be released automatically and the customer told.
A slow source system. If availability comes from a booking system that returns in three seconds at peak, the conversation breaks. On the voice channel latency lands directly on experience; we cover this in the latency budget article. Pre-fetching availability or caching it briefly usually solves it on its own.
How people say dates and times
Customers do not state appointment dates in calendar format. Real recordings contain phrasings like "next Tuesday", "sometime in the afternoon next week", "the 15th", "Saturday, early", "around three", "late afternoon".
Resolving these is a two-layer job. The first layer turns a relative expression into a date, and here the reference date and time zone have to be right. A system that does not account for "next Tuesday" resolving differently depending on today's weekday runs with a one-day skew, and that error propagates silently.
The second layer manages ambiguity. "Around three" is probably 15:00, but you need certainty that it is not 03:00. "Early morning" means different things at different organisations. When it stays ambiguous, the assistant should ask one clarifying question rather than guess.
Read-back is not negotiable
Repeating the date, time, service type and party size back to the customer before creating the record prevents most incorrect bookings. It is a long-standing practice in critical operations and costs a few seconds.
In deployments without a read-back step, the most common failure is a correctly understood date written to the wrong month, and it only surfaces when the customer arrives.
Where reminder calls sit legally
Reminder and confirmation calls relate to an ongoing service relationship and are assessed as transactional notifications under the commercial messaging exemptions in Turkey. The condition attached to that exemption is that the notification must not promote any product or service.
Adding "by the way, we have an offer this month" to the end of the same call makes it a commercial message. That boundary has to be enforced structurally in the flow; the details are in the outbound calling compliance article.
What changes by sector
In healthcare appointments are typically booked far in advance, so two-stage reminders and waiting-list refill deliver the highest return.
In hotels and accommodation, cancellation terms are contractual; the assistant conveying the policy correctly matters for both satisfaction and dispute handling.
In restaurants the booking window is short, the reminder goes out the same day and the time available to refill is very tight. The gain there lies in learning about a cancellation even an hour ahead.
In field service the customer has to be present at the address; a confirmation call prevents wasted journeys as well as missed appointments.
Metrics to track
- No-show rate, on the definition you fixed
- Late cancellation rate and early cancellation rate, separately
- Refill rate for released slots, and time to refill
- Reminder reach rate, meaning reached rather than dialled
- Reschedule rate after a reminder
- Share of records corrected during read-back
The last one looks minor but reports directly on flow quality. A high correction rate means something in the understanding layer is off.
Deployment checklist
- Are no-show, cancellation and organisation-side cancellation defined in writing, and is the denominator explicit
- Is the reminder flow going live with a control group
- Are there two reminder stages with different purposes
- Is cancellation or rescheduling among the first options in the reminder
- Is the slot hold window defined, and released when it expires
- Are relative date expressions and time zone resolved correctly
- Is there a read-back step before the record is created
- Is promotional content kept out of the call
To see how appointment and reservation flows are built end to end, look at Reservation Systems.
References
- Gurol-Urganci, I. et al. (2013). Mobile phone messaging reminders for attendance at healthcare appointments. Cochrane Database of Systematic Reviews. cochranelibrary.com
- Hallsworth, M. et al. (2015). Stating Appointment Costs in SMS Reminders Reduces Missed Hospital Appointments: Findings from Two Randomised Controlled Trials. PLOS ONE. pubmed.ncbi.nlm.nih.gov
- Republic of Türkiye Ministry of Trade, Message Management System (İYS). ticaret.gov.tr