Can AI reduce patient no-shows? What the evidence on reminders says

Every practice owner I know has a no-show number they carry around in their head, and most of them have been pitched an AI tool that promises to fix it. Before you buy the prediction engine, it is worth knowing what the research actually supports. The strongest evidence is older and plainer than the pitch.

Do appointment reminders reduce no-shows?

Yes, with moderate-quality evidence behind it. A 2013 Cochrane review of eight randomized trials with 6,615 participants found that text message reminders improved attendance compared with no reminder. Across those studies, attendance was 67.8% with no reminder, 78.6% with text reminders and 80.3% with phone call reminders.

Read those percentages as what happened in those trials, not as a forecast for your practice. The review authors rated the evidence low to moderate, noted the studies differed a lot from each other, and said the evidence was still not enough to settle policy on its own. The direction is clear. The size of the effect in your clinic is something you measure.

Are text reminders as good as phone calls?

In the Cochrane review, about as good, and cheaper. Three studies comparing text with phone call reminders found a similar effect on attendance, and two studies reported that the cost per attendance was 55% and 65% lower for texts. For most practices that makes text the default, with a call kept for patients who ask for one.

This is where a lot of AI tools are really doing their work: sending the reminder, reading the reply, and moving or cancelling the slot when a patient answers. That is useful automation. It does not need to be sold as prediction to be worth paying for.

The review found almost nothing on harms. None of the included studies reported in detail on problems such as loss of privacy or messages that never arrived. For a practice those are the failures that matter: a reminder sent to an old number, or a text that names the visit type on a shared phone. Nobody measured them, so you have to design against them yourself.

Can AI predict which patients will miss their appointment?

Partly, and less reliably than vendors suggest. A 2020 systematic review of 50 no-show prediction studies found logistic regression was the most common method, and only one study reached an area under the curve above 0.9. The authors called no-show prediction a challenging and unsolved problem. The strongest single predictor was the patient’s own history of missed appointments.

That last finding is useful by itself. You can pull a list of patients with two or more missed visits from your practice management system this afternoon, without any model. Other predictors the review found across studies included lead time, age, day of the week and distance, which your scheduler probably already knows by instinct.

Where does AI actually help with no-shows?

In the handling around the reminder. An agent can send it at the time your patients respond to, read a reply like “can I come Thursday instead,” offer real open slots, and fill a cancelled slot from a waitlist before it goes empty. Those are rule-shaped tasks with a clear record, so every one can be checked afterwards.

1. Measure your current no-show rate by provider and visit type for the last three months.
2. Make sure every patient gets a text reminder, and that replies go somewhere a person or agent reads the same day.
3. Flag patients with a history of missed visits and give them a second reminder or a personal call.
4. Add waitlist backfill so a cancellation becomes a filled slot instead of an empty one.
5. Only then test a prediction tool, against the same three-month baseline.

Keep a person on the exceptions. A patient who replies to a reminder with a question about symptoms or medication has not sent a scheduling message, and the agent should pass that reply to clinical staff the same day instead of answering it.

What can a reminder say without breaking HIPAA?

HHS treats appointment reminders as part of treatment, so they can be sent without a patient’s authorization. For voicemails, HHS says practices should limit what they disclose, and suggests leaving only the practice name, a number and what is needed to confirm, or asking the patient to call back. Keep the visit reason out of the message.

If a vendor sends the reminders for you, it is handling patient information on your behalf, and needs a business associate agreement. And if you move to automated voice calls with an AI-generated voice, the FCC’s 2024 ruling brings them under the consent rules for artificial voices. Check both with counsel before you switch anything on.

What this evidence does not tell you

The Cochrane trials are more than a decade old, and none of them tested an AI system. The prediction review looked at research models, not the products on the market today. Nobody has published a trial that tells you what a specific vendor will do for your no-show rate. So set a baseline, change one thing at a time, and let your own numbers decide.

A useful baseline is simple. Count booked visits and missed visits by week for the last three months, split by provider and visit type. Change one thing, such as reminder timing or waitlist backfill, and compare the next three months against the same count. If a vendor cannot show you results in that shape, it cannot show you results.