Predicting No-Shows for Appointments and Bookings

A missed appointment is a cost you already paid for

A physiotherapy clinic with six practitioners, each seeing 14 patients a day, loses a slot every time someone does not turn up. The practitioner's time, the room and the reception work are all still paid for. At even a modest no-show rate, that adds up to several practitioner-days a month of capacity nobody used.

Clinics, salons, garages, driving schools, restaurants, tutoring services and field service companies all share this problem. And almost all of them handle it the same way: identical reminders to everyone, and a shrug when the slot sits empty.

What predicts a no-show

No-shows are more predictable than they feel. The patterns tend to be similar across industries.

• Past attendance. Someone who missed two of their last five bookings is the strongest single signal.
• Lead time. Bookings made weeks in advance are missed far more often than those made for tomorrow.
• New versus returning customer. First appointments are missed more often, especially if free.
• Payment or deposit. No money down usually means more no-shows.
• Time and day. Early mornings, Monday slots and the day after bank holidays often suffer.
• Reminder response. No reply to a confirmation request is a strong late signal.
• Booking channel. Online bookings made at midnight behave differently from ones arranged at the front desk.

Weather and travel distance help for some businesses. For most, the first four signals carry the bulk of the predictive power, and a simple model captures them well.

What to do with a no-show risk score

Making it easy to cancel matters as much as reminding. Many no-shows are people who meant to cancel but found it awkward. A cancellation two days ahead is a slot you can refill; a no-show is not. Reminder and rebooking flows are routine automation work, and our post on AI appointment scheduling automation covers the booking side in more detail.

The overbooking trap

Airlines overbook, so why not clinics? Because an airline passenger bumped to a later flight is compensated and moved on. A patient who turns up and waits 90 minutes because the model double-booked their slot has a bad experience, and in healthcare may miss care they needed.

There is also a fairness problem. No-show risk often correlates with things like income, shift work, caring responsibilities, disability or reliance on public transport. If high-risk patients are routinely given overbooked slots, the people with the hardest lives get the worst service. In healthcare settings in particular, that is a result we would not want to build, and many organisations have explicit obligations to avoid it.

Use the score to give high-risk bookings more support, not a worse slot.

Where overbooking is acceptable, such as a restaurant or a tyre fitting centre with flexible bays, keep it modest and monitor waiting times as closely as utilisation.

An illustrative example

Take a dental practice group with five surgeries, 1,000 appointments a week and a no-show rate that everyone agrees is too high. A model trained on two years of bookings might put 12% of appointments in a high-risk band containing a large share of the eventual no-shows. Personal calls to that band alone, around 120 calls a week, is a manageable job for reception. If those calls turn even a third of likely no-shows into confirmations or early cancellations that get refilled, the recovered capacity is substantial. The figures are illustrative; the approach of concentrating human effort on a small, high-risk band is what makes it work.

When a model is unnecessary

1. Low volume: a sole practitioner can see who often misses and act on it
2. Deposits already solve it: where a booking fee is normal and accepted, no-shows may already be rare
3. No capacity to act: if nobody can make calls or run a waiting list, a score changes nothing
4. Booking systems with built-in risk flags that already do a reasonable job

Try simple rules first too. 'Missed one of the last three appointments' as a flag gets surprisingly far. A model earns its place when rules leave too many false alarms for staff to follow up.

How we build it

At SpiderHunts, a no-show project usually has three parts: a model scoring each booking nightly, integration with the booking system and messaging tools, and a waiting-list or refill flow for released slots. The model is the smallest part. Healthcare clients also get a data protection impact assessment and a fairness check of who ends up in each risk band before anything goes live. If a conversational assistant for confirmations and rescheduling would help, our chatbot development team builds those on top of the same scores.

Frequently asked questions