Automated Recall and Reminder Messaging That Patients Answer

Most practices already send recall messages. Very few send recall messages that convert, and the gap between those two states is worth five figures a year to a two-surgery site. The difference isn’t the software. It’s the sequence, the channel, the wording, and whether anyone is watching the four numbers that tell you it’s working.

This page deals with one narrow question: how do you set up automated dental recall reminders in a UK practice so that a meaningful share of patients actually book? If you want the wider picture of how recall fits alongside triage, call handling and online booking, the front desk and recall automation pillar covers that ground. What follows goes a layer down.

Recall is a list problem, not a letter problem

Since the 2006 GDS contract, NHS patients are not registered with a practice. There is no list held by the commissioner that you can draw down. Your recall list is something you built and maintain yourself, inside SOE EXACT, Carestream R4, Dentally, Systems for Dentists or whatever else you run, and it decays quietly.

Decay looks like this. A patient moves house and the mobile number stays live but the email bounces. Someone gets recalled at 6 months when their NICE CG19 interval was set to 12. A child turns 18 and drops off a family reminder that was keyed to the parent’s record. A lapsed patient from 2022 is still counted as “active” in your head but hasn’t been seen in 38 months.

Before automating anything, pull a list of patients with a next-recall date in the past 24 months and sort by last visit. On a 4,500-patient database, expect somewhere between 12% and 25% of those records to have a dead or duplicated contact route. Automation applied to a dirty list produces confident-looking dashboards and no bookings.

The four numbers that tell you it works

“Response rate” is not one number, and vendors quote whichever one flatters them.

The last row is the one that matters and the one nobody measures, because it needs attribution across four touches and two channels. If a supplier can only show you opens, they are selling you a mail merge.

The sequence, not the message

A single text on the due date is the most common setup in UK practice and the weakest. Patients read it, intend to ring, and don’t. The fix is a staged sequence with different channels and escalating specificity.

Day  0   SMS      Recall due, one-tap booking link
Day  7   Email    Longer form, includes what the appointment covers
Day 21   SMS      Names a specific slot held for 48 hours
Day 30   Task     Drops onto reception's call list with full context
Day 60   Email    Final, then record moves to "lapsed" cohort

Day 21 does the heavy lifting. A message that says “we can see you Tuesday 14th at 3.40pm, reply YES to take it” converts roughly twice as well as one that says “please call to book”, because it removes the phone call, the hold music and the decision about when.

The Day 30 task is the part people delete to save money, and it’s the part that rescues the highest-value patients. A Band 3 case or a private crown review is worth a two-minute call. Reception should get the record with the last visit date, the treatment history and the reason for recall already on screen, not just a name and a number.

Worked example: 4,500 active patients

Take a mixed NHS and private practice, three surgeries, 4,500 patients on an average 8-month recall. That’s roughly 560 patients falling due each month.

Under a single-text setup at a 41% booking rate, 230 book. Under the five-step sequence at 68%, 381 book. The delta is 151 additional attended exams a month.

On the NHS side, a Band 1 exam is 1 UDA. At a £30 UDA value that’s £30 of activity. Assume 22% of those exams generate Band 2 treatment (3 UDAs, £90) and 4% generate Band 2b (5 UDAs, £150). The arithmetic on 151 extra exams:

151 exams        × £30   = £4,530
33 Band 2 cases  × £90   = £2,970
6 Band 2b cases  × £150  =   £900
                           -------
                           £8,400 per month of additional activity

Messaging cost for the same month: 560 patients receiving an average of 2.3 messages is about 1,290 sends. At 3.5p per SMS segment and effectively nothing for email, that’s around £45. The platform licence sits somewhere between £70 and £250 a month depending on supplier and seat count.

One honest caveat. If you are already delivering 100% of your contracted UDAs, extra NHS activity has no financial value and may actively hurt you by consuming chair time you needed for private work. The case for recall automation on a full contract is about clawback protection at the other end of the year, patient retention, and keeping the private book full. Work out which situation you’re in before you model the return.

Channel economics, and the curly apostrophe problem

SMS in the UK runs roughly 2.5p to 4.5p per segment on volume-tier accounts with providers like Esendex, ClickSend or Textlocal, and a little more through Twilio if you’re building something custom. A segment is 160 characters using the GSM-7 alphabet.

Here’s the trap. Drafting your template in Word and pasting it in converts straight apostrophes to curly ones. A curly apostrophe is not in GSM-7, so the whole message silently switches to UCS-2 encoding, and the segment length drops from 160 characters to 70. A 145-character recall text that cost you one segment now costs three. Across 6,700 sends a year that is the difference between £235 and £700, for one invisible character. Emoji do the same thing. Type your templates in a plain text editor and check the segment count before you go live.

WhatsApp Business is tempting because read rates are higher, but it is priced per template conversation rather than per segment, and utility-category templates in the UK are in a similar ballpark per message once you account for the 24-hour window. It’s worth it for practices with a large cohort who don’t answer SMS, less so as a blanket replacement.

Email costs nothing and converts poorly on its own. Keep it as the supporting channel that carries detail, not the one that asks for the booking.

Writing the thing so it gets a reply

Reception staff write recall texts the way they’d write a letter. Patients read them the way they read a bank alert: in two seconds, on a lock screen, deciding whether to open.

Three rules hold up under testing. Put the practice name first, because an unrecognised number is deleted. Give one action, not two. Never include a marketing line in a recall message, for reasons covered in the next section.

Marshall St Dental: Hi Anna, your check-up is due.
Book online: msd.uk/b/4471 or call 0161 xxx xxxx.
Reply STOP to opt out.
                                    [148 chars, 1 segment]

Compare with the Day 21 version:

Marshall St Dental: Hi Anna, we've held Tue 14 Oct,
3.40pm for your check-up. Reply YES to confirm or
NO and we'll offer it elsewhere.
                                    [137 chars, 1 segment]

The second one gets replies because it costs the patient one keystroke and creates a small, polite deadline.

NHS specifics that change the setup

The recall interval box on the FP17 is not decoration. Setting it per NICE CG19 (3 to 24 months for adults, 3 to 12 months for under-18s) and then driving your automation from that field rather than a blanket six-month default is the single biggest accuracy improvement available. It also means your recall pattern is defensible if anyone asks why a high-risk patient was on 3 months and a low-risk one on 24.

Practices delivering between 96% and 100% of contracted UDAs keep full payment. Fall below 96% and the commissioner recovers the shortfall. Recall automation is a clawback instrument: run a mid-year report in October, work out your run rate, and if you’re tracking at 88%, that is the moment to fire a targeted campaign at lapsed Band 1 patients rather than in March when there’s no chair time left.

Consent, PECR and where practices get caught

A reminder that a patient’s check-up is due is a service message about care they have received, not direct marketing. Under UK GDPR your lawful basis is typically Article 6(1)(f) or 6(1)(e) for NHS work, with Article 9(2)(h) covering the health data element. PECR’s consent rules for electronic marketing don’t bite.

Add “and ask about our £199 whitening offer” to the end of that same text and the whole message becomes direct marketing. Now you need consent for that patient, and you need it recorded. This is exactly how practices end up with an ICO complaint from a single annoyed patient over a message they thought was routine.

Keep an opt-out on every send, honour it within the practice management system and not just in the messaging tool, and make sure a patient who opts out of texts still appears on reception’s call list rather than vanishing.

Where AI genuinely adds something

Scheduling the sequence is rules, not intelligence. Any competent platform does it.

The parts where a model earns its licence fee are narrower. Predicting which patients in the due cohort are least likely to respond, so the Day 30 call list gets prioritised by risk rather than alphabetically. Parsing free-text replies, because a real inbox contains “yes but can we do after 4”, “my husband has moved out”, and “I’ve changed dentist”, and a keyword matcher handles one of those three. Handling the overflow voice calls that a recall campaign generates on Monday morning, which is the failure mode nobody models: send 560 texts, get 90 phone calls, and watch your answer rate collapse on the same day you spent money trying to reach people.

Run the campaign in three tranches across a fortnight rather than one batch, and check what your call answer rate does on the day after each send.