An unanswered reminder barely moves the number
Reminders are standard practice for any business that books time. The reminders go out, the slots still empty, and the conclusion drawn is that reminders do not work very well. That conclusion is nearly right and completely useless. Reminders work. Reminders nobody answers do not, and a reminder programme is usually measured as though those two things were the same event.
A 2018 study published in Psychiatric Services by the American Psychiatric Association split reminder-call outcomes by what actually happened on the line, and found that 39% of patients did not turn up where the reminder call went unanswered, against 3% where the call turned into a real conversation. Same reminder, two outcomes. The variable was whether anyone answered. The setting was clinical, and those two rates belong to that setting. What carries over to a salon or a garage is the mechanism, not the percentages.
39% no-show when the reminder call goes unanswered, 3% when the reminder becomes a conversation. (Psychiatric Services, APA, 2018)
That gap is wider than anything a rewritten script or a different send hour will ever produce. A reminder is normally treated as a notification: information pushed at somebody, delivery assumed, task closed. The finding says the value was never in the delivery. It was in the exchange.
When the customer answers and says yes, two things happen that a delivered message cannot do. The customer commits out loud about a specific hour, and you learn something you did not know before the call. A silent reminder gives you neither. You end up with the same blind calendar you had that morning, plus the comfortable belief that the risk has been covered.
Why SMS and voicemail lose on the same logic
Stated that way, the weakness of the cheap channels stops being a matter of taste. The problem is the direction of travel. SMS, email and voicemail deliver, and they rarely return an answer, which leaves you standing on the unanswered side of the Psychiatric Services split.
Voicemail is the clearest case. An industry estimate from BIA/Kelsey puts at 85% the share of people who reach voicemail and never call back, which tells you how much intent survives a machine picking up. The same reluctance shows on your own line, where CallRail's 2025 inbound data found that 42% of callers leave a voicemail. Leaving a reminder on an answering machine means settling for a delivery receipt instead of a confirmation.
Then there is the unknown number. Arcep, the French telecoms regulator, reported in 2024 that 65% of consumers in France rarely or never answer a call from a number they do not recognise, up from 57% two years earlier. That is one market and one regulator, and we have no equivalent figure for yours. It still names the habit a reminder call has to work against: an unidentified ring gets treated as spam until proven otherwise. A reminder placed from an anonymous number was built to land on the losing side of the study.
None of that makes SMS useless. It is a good carrier for detail: the address, the hour, the parking, the document to bring. It is a poor instrument for confirmation, because a message that went unread and one that was read and ignored look identical in your system. A reminder programme running on text alone can post excellent delivery figures and unchanged no-shows at the same time, and there is no contradiction in that.
Delivery is not confirmation
A reminder programme that reports delivery rates is reporting how well the phone network functions, not how many slots it saved.
The reminder is a confirmation, and confirmation runs both ways
So the design brief for a reminder call is narrow and slightly awkward. It has to reach somebody who did not want to be interrupted, hold them for well under a minute, and come back with one of three answers: coming, not coming, or needs a different time. Anything short of one of those three and the call was decoration with a phone bill attached.
The third answer is the one most reminder systems discard. A reminder that can only accept yes or no has a built-in failure mode: the customer who wants to come but cannot make that hour. Told only that the appointment exists, they do the easiest available thing, which is nothing, and the slot dies on the day. Given a concrete alternative inside the same call, the same person becomes a rebooking. The slot moves instead of disappearing.
A reminder that cannot reschedule on the spot converts the customer's small scheduling problem into your empty hour.
This is where an AI voice agent earns its place, and it is worth being exact about why. Not because it sounds convincing. Because it can hold these short two-way exchanges in parallel, as many at once as tomorrow's list needs, and each one can end in a decision rather than a delivery. A live team does the same work better, one call at a time, which is precisely why the reminder list never gets staffed. It is low-status work whose success looks like nothing happening.
Vendor-reported industry data for 2025 puts the median no-show reduction from AI reminder systems at around 34%, and that number deserves the hedge it carries: it is compiled from suppliers describing their own deployments, across businesses with unlike baselines. Treat it as an order of magnitude. The Psychiatric Services split is the finding to plan against, because it explains the mechanism instead of averaging outcomes.
Where the slot is really lost: the call in the other direction
Reminders get the attention because they are the part you control. The larger leak usually runs the other way, on the inbound call from the customer who has already decided to be helpful and is ringing to move the appointment rather than skip it. That call is a gift. It arrives, and then it queues.
ContactBabel's 2024 figures for UK contact centres put the average wait before a caller hears a human voice at 116 seconds, and found that 8.4% of inbound callers hang up before anyone answers. A 2012 Velaro survey found around 60% of callers will not hold longer than one minute. The customer trying to reschedule is doing you a favour, and the favour expires inside a minute.
Smaller operations fare worse, because there is no queue at all, only a phone nobody can get to. A 411 Locals study in 2016 found 62% of calls to small businesses went unanswered. Anyone who has run a clinic reception, a salon or a garage knows the mechanism without the study: the person who answers the phone is the same person serving the customer standing in front of them.
The customer who calls to move an appointment has already decided not to waste your slot. Then they hold.
Which is why the outbound reminder and the answered inbound line are one system rather than two projects. The reminder catches the people who forgot. The answered line catches the people who remembered and have a problem. Automate one, neglect the other, and you will have moved your no-shows into your missed calls and reported it as progress.
What the agent needs access to before the call is worth placing
An agent that can read the appointment out loud and do nothing else is a text message that costs more. To come back with one of the three answers, and specifically to handle the third one, it needs real access to the systems that own the calendar. This is where reminder projects actually fail. The conversation design is the easy half.
- Write access to the calendar, not read access. An agent that can see the booking but not change it hands every reschedule back to a human, which returns the slot to the queue it was trying to escape.
- The real availability rules. Which room, chair, bay, vehicle or specialist the appointment consumes, how long it takes, what cannot sit next to what, how much notice a change requires. Availability is rarely just an empty box in a grid.
- The cancellation and rebooking policy, including any fee window, so the agent states the consequence accurately instead of improvising it.
- Enough identity verification to discuss the booking at all, at whatever level your sector requires, and a hard refusal to discuss details when the check does not pass.
- A record written back the moment the call ends: what was decided, the new time, the reason given. A reschedule that exists only in the audio is not a reschedule.
- A clean handoff to a person for anything outside the reminder's scope, carrying the context across, so the customer is not asked to start again from the beginning.
The order of work matters. Get the integrations wrong and a beautifully written agent will hold a pleasant conversation and then fail to act on it, which is worse than never calling: you asked the customer to commit, then lost the commitment.
How to place the call so that it gets answered
None of this helps if the phone is not picked up. Our founders spent about ten years running live call centres, up to 1,500 operators at peak across France, Mexico and beyond, and the rules that made an outbound list connect have not changed because the caller is now software.
- Call from a number the customer recognises and can call back, ideally the same number they booked on. Given how an unrecognised number gets treated, caller identity is the thing to fix first, and it is usually a configuration setting rather than a project.
- Say who you are and what this is about in the opening sentence. The customer decides whether to stay on the line almost immediately, and naming the appointment and its hour does more work than any greeting.
- Ask a closed question that requires an answer. Confirming is a decision. Saying you look forward to seeing them is not a question and returns no data.
- Have the alternative ready before you need it. When the answer is not that hour, the next sentence should be concrete open slots, not an offer to call back later.
- Retry differently rather than repeatedly. Vary the hour of the attempt, and let the cheap channel carry the detail while the call carries the confirmation.
- Measure confirmations, not contacts. Attempts, deliveries and answered calls are all easy to chart, and none of them is the number that pays for the work.
What all of it produces is one figure most calendars never show: how many of tomorrow's bookings have been confirmed out loud by the person who made them. Not reminded. Confirmed. Once that figure exists on a screen, the argument about channels settles itself, because only one channel supplies it.
One place automation loses, worth naming rather than hiding. A reminder tied to bad news, a difficult result or a legal deadline is not routine work, and an agent that handles it as routine will do damage no recovered slot pays for. Mark those lists and let a person make those calls. The rest is short two-way exchanges, and that is the work a neuro-agent is built for.