Key takeaways
- Automation is good at the predictable parts of a call: opening hours, directions, booking or moving routine appointments and taking messages.
- Anything clinical, distressed, complex or about a complaint should reach a person quickly, with a clear escalation rule.
- Software that triages or advises on symptoms may be a regulated medical device; the MHRA says many health software and AI products are[1].
- Ask suppliers for evidence on escalation, data handling and failure modes before you look at features.
What automated phone booking means today
The phrase covers several quite different things. A keypad menu that routes calls ("press 1 for appointments") is automation. So is a system that recognises speech and lets callers say what they want. The newest category is the AI voice assistant, which holds a conversation, answers questions and can take actions such as offering appointment times or passing a message to staff.
For private clinics, the appeal is clear. Calls cluster at opening time and lunchtime, staff are busy with patients at the desk, and every unanswered call may be a patient who books elsewhere. Ease of booking matters to private patients: 28% of people who have gone private say they did so because it is easier to book appointments[2].
Patients are increasingly used to talking to software. In the government's public attitudes tracker, 40% of UK adults said they use AI chatbots for personal purposes at least once a month, up from 34% in the previous wave[3].
Source: GOV.UK (Department for Science, Innovation and Technology), 2024[3]
Show the data
| Item | Value |
|---|---|
| Previous wave | 34% |
| Wave 4 | 40% |
What automation handles well, and what needs a person
| Call type | Automation | Person |
|---|---|---|
| Opening hours, address, parking, directions | Handles fully from approved wording | Only if the caller asks for a person |
| Booking a routine follow-up with a known clinician | Can offer times and book where integrated with the diary | When the caller is unsure what they need |
| Moving or cancelling an appointment | Handles well, including freeing the slot | If the change affects a treatment plan |
| New patient enquiry about services | Explains services and how to book | Questions about suitability or clinical detail |
| Results, symptoms or medication questions | Takes a message; does not interpret | Clinician or trained staff respond |
| Distressed or confused caller | Recognises and transfers | Always |
| Complaint | Takes details and transfers or books a call-back | Always handled by a person |
| Possible emergency | Gives 999 or NHS 111 wording immediately | Staff informed afterwards |
The pattern is consistent: automation for information and simple transactions, people for judgement, emotion and anything clinical.
The escalation rules that must exist
Before any automated system answers a single patient call, write down when it must hand over and how. Without explicit rules, systems either hand over too often, which defeats the purpose, or too rarely, which is unsafe.
- Emergency words: if a caller mentions chest pain, difficulty breathing, signs of a stroke, heavy bleeding, suicidal thoughts or similar, the system gives 999 or NHS 111 wording straight away. Agree the exact wording with a clinician.
- Clinical questions: any question about symptoms, results, medicines or whether treatment is needed goes to a person or becomes a message for clinical review.
- Caller asks for a person: transfer, or book a call-back within a stated time if no one is free. Never trap callers in a loop.
- Distress or confusion: transfer to a person as soon as the system detects it, or after a failed second attempt to understand.
- Safeguarding concerns: transfer immediately and follow the practice safeguarding process.
- System failure: if the system cannot complete a task, it says so and offers a person or a call-back.
Test each rule with real staff role-playing callers before going live, and review a sample of escalated calls every week for the first months.
Where the regulatory lines are
Three areas need checking before you buy.
- Medical device regulation: the MHRA says many software and AI products used in health and social care are regulated in the UK as medical devices[1]. A system that answers practice questions and books routine appointments is different from one that assesses symptoms and advises on urgency. If a product claims to triage or advise, ask the supplier for its regulatory status.
- Data protection: calls about health appointments often involve health data, which is special category data under UK GDPR Article 9[4]. You need a data protection impact assessment, a data processing agreement with the supplier and clear information for patients about how calls are handled and recorded.
- Recording and transcripts: if calls are recorded or transcribed, tell callers at the start, set retention periods and control who can access them.
Patients care about these points. Healthcare is the area where UK adults most often say AI needs careful management to prevent negative outcomes[5]. Being clear about what the system does and does not do is part of earning that trust.
Questions to ask suppliers
- Which call types does the system handle end to end, and which does it only take a message for?
- How does it decide to escalate, and can we see and change the escalation rules?
- What happens when it does not understand the caller, and after how many attempts does it hand over?
- Can callers always reach a person, and how quickly?
- Does it write appointments into our booking system directly, or create requests for staff to confirm?
- Where are call recordings and transcripts stored, for how long, and who can access them?
- What is the product's regulatory status if it gives any advice about symptoms?
- Can we approve every piece of wording it uses about our services, hours and urgent care?
- What reporting do we get on calls handled, transferred and abandoned?
- How are updates tested, and will we be told before behaviour changes?
Ask for a pilot on a limited set of call types, with clear exit terms if it does not perform.
Prepare your information before go-live
An automated system is only as accurate as the information it is given. Most problems in the first weeks come from out-of-date or incomplete practice details, not from the technology. Before switching anything on, assemble one approved source of truth.
- Opening hours for each site, including bank holidays and any lunchtime closures.
- Services offered at each site, described in plain words, with what each one involves and who it is for.
- Clinicians, their days and the appointment types each one offers.
- Directions, parking, public transport and step-free access for each site.
- Cancellation terms and how patients can change an appointment.
- The exact wording for emergencies and urgent care, signed off by a clinician.
- What the system should say when it does not know the answer.
Name one person as the owner of this information and agree how changes reach the system, for example within a working day of a clinician changing their days.
Getting the greeting right
The first ten seconds decide whether callers trust the system. A good greeting says who is answering, what it can help with and how to reach a person. Example: "Thank you for calling Riverside Clinic. I'm the clinic's automated assistant. I can help with appointments, directions and opening times. If you need to speak to the team, just say so. If this is an emergency, please hang up and call 999."
Keep it short and avoid pretending to be a person. Callers who feel misled are more likely to complain, even if the call itself went well. Repeat the route to a person at any point where the caller hesitates.
Measures to judge performance
Track the same figures before and after introduction so you can see the effect.
- Time to answer, and the share of calls answered within a minute.
- Calls abandoned, especially at menu or greeting stages.
- Share of calls completed without a transfer, by call type.
- Escalations, and a weekly sample reviewed for whether each one was right.
- Complaints or comments mentioning the phone system.
National benchmarks give a sense of scale. NHS 111's average time to answer in July 2026 was 82 seconds, with 75.6% of calls answered within 60 seconds[6], and 4.6% of calls were abandoned after the call-steering menu[7]. A clinic line is very different from a national service, but the figures show that even well-run telephone services lose some callers at the menu stage.
Example: a phased introduction
Example, fictional: a private clinic group with three sites introduced an automated phone assistant in three phases. Phase one handled only information calls: hours, directions, parking and services. Phase two added moving and cancelling existing appointments. Phase three added booking routine follow-ups for known patients. New patient bookings and anything clinical stayed with reception throughout. At each phase, the practice manager listened to 20 calls a week and checked every escalation. Phase three was paused for a month when the system offered follow-ups with clinicians who had changed their days, and resumed once the diary integration was fixed.
Where an AI Employee fits
Some products go beyond booking to cover a wider range of front-desk calls. Ava, for example, answers practice questions in wording the practice approves, explains how to get help, gives agreed 999 and NHS 111 wording and passes anything about a patient's care to the reception team. Writing appointment requests directly into clinical systems is a planned capability rather than a live one, because it needs clinical safety assurance and a data protection review first. Whatever you choose, judge it on the escalation rules and the evidence from your own pilot, not on a feature list.
Keep people in the loop
Automation works best when staff can see what it is doing. Make sure reception can read call summaries, pick up messages quickly and correct any wording that confuses callers. Tell patients what to expect in your phone greeting and on your website, including how to reach a person. Review the system every quarter, just as you would review a member of staff's work.
Frequently asked questions
Can automated phone systems book appointments directly?
Some can, where they are integrated with the booking system. Others create requests for staff to confirm. Ask suppliers exactly which they do for each appointment type.
Is an AI phone assistant a medical device?
It depends on what it does. The MHRA says many health software and AI products are regulated as medical devices. Answering practice questions and booking routine appointments is different from assessing symptoms; ask suppliers for their regulatory status.
Will patients accept talking to an automated system?
Many will for simple tasks, especially if they can reach a person easily. Make the route to a person clear and test the experience with real callers.
What should never be automated?
Clinical judgement, complaints, safeguarding concerns and conversations with distressed callers should always reach a person.
References
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2025). Software and artificial intelligence (AI) as a medical device (opens in a new tab) ↑
Independent Healthcare Providers Network (IHPN) (2026). Support for using private healthcare now found across the political spectrum, new report shows (opens in a new tab) ↑
GOV.UK (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 9: Processing of special categories of personal data (opens in a new tab) ↑
GOV.UK (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
NHS England (2026). Statistical Note: Integrated Urgent Care Aggregate Data Collection (IUC ADC) July 2026 (opens in a new tab) ↑
NHS England (2026). Statistical Note: Integrated Urgent Care Aggregate Data Collection (IUC ADC) July 2026 (opens in a new tab) ↑