AI voice agent developments handle inbound patient calls for UK medical and dental clinics 24 hours a day, 7 days a week - booking appointments, answering opening hours queries, managing prescription renewal requests, and routing urgent calls to on-call staff. For a GP practice or dental clinic receiving 200-400 calls per week, an AI voice agent reduces receptionist call-handling time by 55-70% and eliminates missed calls outside opening hours. Implementation costs £3,000-£8,000 and takes 4-8 weeks. Softomate Solutions deploys AI voice agents for clinics on VAPI and ElevenLabs infrastructure, integrating with EMIS Web, SystmOne, Dentally, and Software of Excellence.
Last updated: 18 May 2026
Published 18 May 2026The majority of inbound calls to a GP surgery or dental practice are transactional in nature. They follow predictable patterns: a patient wants an appointment, wants to know the surgery's opening times, needs to request a repeat prescription, or has a question about a recent test result. Research from NHS England and primary care network data consistently shows that between 60% and 75% of all inbound calls require no clinical judgement whatsoever - they are administrative tasks that can be handled by a well-configured AI voice agent without any compromise to patient safety or experience.
A modern AI voice agent deployed in a UK clinic context operates as the first point of contact on the phone line. The patient calls the clinic number, the AI answers within one or two rings, identifies the patient using a standard date-of-birth or postcode verification check, and then handles the enquiry from start to finish. If the call requires a clinician - an urgent triage, a prescription query requiring GP sign-off, or a complaint - the AI escalates seamlessly to the relevant staff member or on-call team, with a full transcript of the conversation handed over in real time.
Below is a breakdown of the call types an AI voice agent can handle autonomously versus those that require human escalation:
| Call Type | AI-Handled (Autonomous) | Human Escalation Required |
|---|---|---|
| Appointment booking (routine) | Yes - checks live availability and books | No |
| Appointment cancellation or rescheduling | Yes - updates the clinical system | No |
| Opening hours and location queries | Yes - configured knowledge base | No |
| Repeat prescription routing | Yes - logs request, notifies relevant staff | No (GP reviews separately) |
| Test result status enquiry | Partial - signposts to online portal or advises timeframe | Yes - if result is abnormal or requires explanation |
| Appointment reminder (outbound) | Yes - proactive outbound call campaign | No |
| DNA (Did Not Attend) follow-up | Yes - calls patient, offers rebooking | No |
| Urgent triage or emergency | No - immediately signposts to 111 or 999 | Yes - escalates to on-call clinician |
| Complaint or dissatisfied patient | No - flags for practice manager | Yes - immediate warm transfer |
| NHS registration enquiry | Yes - provides process information | No |
| Prescription query requiring clinical input | No - logs and routes to pharmacist or GP | Yes |
| Home visit request | Partial - collects details, routes to duty GP | Yes - clinical decision required |
For dental clinics specifically, the AI voice agent adds another layer of value: new patient registration. A prospective patient calling to register with a practice can be walked through the registration form verbally, with the AI capturing the necessary fields and creating a draft record in Dentally or Software of Excellence ready for staff review. This is particularly valuable for NHS dental practices in high-demand areas such as East London, where new patient calls can consume 20-30 minutes of receptionist time per enquiry.
The AI also handles appointment reminder calls proactively. In a practice with a 10-12% DNA rate - the national average for NHS GP surgeries - an automated outbound call 24 hours before the appointment, with the option to confirm or cancel and rebook, consistently reduces DNA rates to 4-6%. That recovery directly improves QOF (Quality and Outcomes Framework) performance and reduces wasted clinical time. For a 10-session-per-day GP practice, recovering even two slots per week represents significant financial and operational value.
Multilingual capability is particularly relevant for urban UK practices. An AI voice agent built on ElevenLabs or VAPI infrastructure can be configured to detect and respond in multiple languages - including Bengali, Gujarati, Punjabi, Urdu, Polish, and Romanian - switching language based on caller preference. For a Barking or Tower Hamlets practice serving a diverse patient population, this removes a significant barrier to access that traditional telephone reception cannot address cost-effectively.
The Care Quality Commission does not prohibit AI voice agents in GP surgeries or dental clinics. What CQC does require - under Regulation 17: Good Governance - is that the provider can demonstrate that any technology used to support patient interaction is safe, effective, and appropriately governed. This means the onus is on the practice to document how the AI voice agent is configured, what it is and is not permitted to do, how escalation works, and how the system is audited over time.
Regulation 17 requires registered providers to maintain accurate, complete, and contemporaneous records of patient interactions. An AI voice agent that does not produce a full, timestamped transcript of every call, accessible to practice management, would fail this requirement. A correctly configured system - as deployed by Softomate Solutions - generates a complete call log, stores it securely in the UK, and makes it available for CQC inspection on request. This is typically more auditable than a human receptionist taking handwritten notes, because the AI record is verbatim and timestamped to the second.
NHS England published its Code of Conduct for Data-Driven Health and Care Technology and has since aligned with the MHRA's Software as a Medical Device framework and the AI and Digital Regulations Service (ADRS). An AI voice agent used purely for administrative call handling - appointment booking, opening hours, prescription routing - does not meet the definition of a Software as a Medical Device under current MHRA guidance, because it does not make or influence clinical decisions. The moment an AI voice agent attempts to interpret symptoms, suggest diagnoses, or recommend treatment pathways, it crosses into regulated device territory. A correctly scoped voice agent avoids this line entirely.
Below is a summary of the regulatory boundaries that govern what an AI voice agent can and cannot do in a CQC-registered healthcare setting:
| Action | Permitted for AI Voice Agent? | Regulatory Basis |
|---|---|---|
| Book, cancel, or reschedule routine appointments | Yes | Administrative - no clinical judgement |
| Collect patient identifiers for verification | Yes - with GDPR-compliant data handling | ICO guidance on special category data |
| Route repeat prescription requests to staff | Yes - routing only, no authorisation | Administrative task; GP retains clinical sign-off |
| Provide opening hours, address, and general information | Yes | No clinical risk |
| Signpost urgent callers to 111 or 999 | Yes - mandatory safety net | CQC Regulation 12: Safe Care |
| Triage symptoms and advise clinical pathway | No | Requires MHRA SaMD classification and clinical validation |
| Advise on medication dosage or interactions | No | Clinical decision - requires registered prescriber |
| Interpret test results for patients | No | Clinical judgement; potential for significant harm |
| Store call recordings in the UK | Yes - required for Regulation 17 compliance | ICO UK GDPR Article 9 (special category data) |
| Transfer call data to third countries | No - without adequate safeguards | ICO international transfer rules |
From a GDPR perspective, patient call data is special category data under Article 9 of the UK GDPR, because it relates to health. This means the practice must have a lawful basis for processing - typically Article 9(2)(h) (health and social care) - and must ensure that data is stored securely, retained only as long as necessary, and not transferred outside the UK without appropriate safeguards. Softomate's deployments use UK-based data residency by default, with call transcripts encrypted at rest and in transit.
Staff training is a CQC compliance requirement, not an optional extra. Reception and administrative staff must understand when the AI will escalate to them, how to access call transcripts, and how to override the system if it is not performing as expected. Softomate includes CQC documentation templates and a staff training session as standard in every healthcare deployment - producing the written policies, escalation procedures, and audit trail that an inspector would expect to see.
NHS dental contracts add a further layer: NHS England's Standard Contract requires practices to maintain continuity of patient access. A voice agent that fails and leaves patients unable to book appointments would breach access standards. For this reason, every Softomate healthcare deployment includes a failover routing rule: if the AI becomes unavailable for any reason, calls are automatically forwarded to a human answering service or voicemail with a callback SLA.
The practical value of an AI voice agent in a clinical setting depends entirely on its ability to read and write to the clinical system in real time. An AI that can only answer pre-scripted questions without checking live appointment availability is not much more useful than an answerphone. A properly integrated AI voice agent reads live slot availability from the clinical system, books the appointment, and sends the patient a confirmation SMS or email - all within a single call lasting under 90 seconds.
EMIS Web and SystmOne together cover approximately 85% of GP practices in England. Both systems expose APIs that allow third-party applications to query appointment slots, patient demographics, and prescription lists - subject to GP systems vendor approval and Information Governance agreements. Softomate works with practices to obtain the necessary API credentials and IG agreements before any integration goes live. The typical IG setup time is 2-3 weeks, which is factored into the 4-8 week implementation timeline.
For dental practices, Dentally and Software of Excellence (SOE) are the dominant clinical management systems. Dentally provides a modern REST API with webhook support, making integration relatively straightforward. SOE's API is more restrictive and typically requires a middleware layer, which Softomate builds as part of the integration package.
Below is a comparison of integration capabilities across the main clinical systems:
| Clinical System | Real-Time Slot Query | Appointment Booking | Patient Matching | Integration Complexity |
|---|---|---|---|---|
| EMIS Web | Yes - via EMIS API | Yes | Yes - DOB + surname | Medium (IG agreement required) |
| SystmOne | Yes - via TPP API | Yes | Yes - DOB + NHS number | Medium (TPP approval required) |
| Dentally | Yes - REST API | Yes | Yes - DOB + postcode | Low (open API, fast setup) |
| Software of Excellence | Partial - middleware required | Yes via middleware | Yes | High (proprietary API) |
| Carestream Dental | Limited | Manual fallback | Yes | High |
| iSmile / Exact | Yes - via Exact API | Yes | Yes | Medium |
Telephony integration is the other critical component. Most UK GP practices and dental clinics already use a VoIP telephone system - typically 3CX, RingCentral, BT Cloud Voice, or Vonage. An AI voice agent integrates at the SIP trunk level, intercepting inbound calls before they reach the receptionist queue. This means no hardware changes are required at the clinic. The AI answers the call, handles or escalates, and only passes to a human extension when the situation demands it.
For practices still on legacy analogue PSTN lines - increasingly rare but not extinct - a SIP conversion gateway is required. Softomate can procure and configure this as part of the implementation package, or work with the practice's existing telephony provider. The typical additional cost for PSTN-to-SIP conversion is £200-£500 for hardware, plus any changes to the monthly line rental.
Patient matching during a call uses a two-factor verification approach: date of birth plus either surname or postcode. This mirrors the access control approach used by NHS 111 and most GP online consultation tools, and meets ICO guidance on telephone identity verification for health data. The AI does not proceed with any appointment booking or prescription routing until it has successfully verified the patient's identity against the clinical system record.
Choosing the right underlying AI voice platform matters for UK healthcare deployments. The three platforms most commonly used for production healthcare voice agents in 2026 are VAPI, ElevenLabs Conversational AI, and Bland.ai. Each has a different profile of strengths and trade-offs relevant to the specific demands of UK clinical environments - particularly around accent quality, data residency, latency, and cost per minute.
VAPI (Voice AI Platform Infrastructure) is a developer-focused orchestration layer that sits between the telephony system and the AI language model. It handles the real-time audio processing, turn-taking, interruption detection, and function calling (the API calls to clinical systems). VAPI is model-agnostic, meaning Softomate can plug in different LLMs and TTS (text-to-speech) engines depending on the requirement. For healthcare, VAPI is typically paired with GPT-4o or Claude 3.5 Sonnet as the reasoning layer, and ElevenLabs voices for the audio output.
ElevenLabs Conversational AI is a more vertically integrated platform: it provides both the voice synthesis and the conversation management layer. ElevenLabs has the most natural-sounding British English voices currently available - including regional accents such as received pronunciation, mild London, and Scottish - which matters for patient trust. Patients calling an NHS GP surgery expect to hear a voice that sounds local and professional. An obviously synthetic or American-accented AI will generate complaints.
Bland.ai is a US-based platform with a strong track record in healthcare call automation in North America. Its UK offering is newer, and data residency options for UK storage are available but require specific configuration. For NHS and CQC-regulated practices, UK data residency is non-negotiable, so Bland.ai requires careful procurement review before deployment.
| Platform | UK Accent Quality | UK Data Residency | NHS Integration | Cost per Minute | Latency (avg) | Multilingual |
|---|---|---|---|---|---|---|
| VAPI (with ElevenLabs voices) | Excellent | Yes - configurable | Yes - via function calling | £0.08-£0.14 | 600-900ms | Yes - 29 languages |
| ElevenLabs Conversational AI | Excellent - best-in-class | Yes - EU/UK region | Yes - via webhooks | £0.10-£0.18 | 700-1,100ms | Yes - 32 languages |
| Bland.ai | Good | Configurable (check SLA) | Partial - custom build required | £0.06-£0.12 | 500-800ms | Yes - 12 languages |
Latency is a particularly important metric for telephone calls in healthcare. A conversation turn that takes more than 1.2 seconds to respond will feel unnatural to a patient who is potentially anxious about their health. VAPI and ElevenLabs both achieve sub-900ms average latency on UK telephony infrastructure, which is within the acceptable threshold for natural conversation. Bland.ai's lower latency figure is attractive, but its more limited multilingual support is a meaningful constraint for East London and other diverse urban practices.
Softomate's default recommendation for NHS-contracted GP practices is VAPI with ElevenLabs voices and UK data residency configured at the outset. For private dental practices with simpler data governance requirements and a desire for the lowest possible per-minute cost, Bland.ai is a viable option subject to a data processing agreement review. For practices requiring the highest voice quality and the widest language support, the ElevenLabs Conversational AI native platform is the strongest choice.
One of the most common questions from practice managers and clinic owners is: what does the return on investment actually look like? The honest answer is that it depends on current call volume, staff costs, and the proportion of calls that are genuinely administrative. The following example uses conservative assumptions based on a single GP surgery in outer East London - a realistic profile for the areas Softomate serves.
A four-GP practice in Barking receiving an average of 320 inbound calls per week employs two full-time receptionists who spend approximately 60% of their working day on the telephone. At a loaded cost (salary plus employer NI plus pension) of £28,000 per receptionist per year, the two-person front desk costs roughly £56,000 annually for call-handling alone. This does not count the cost of missed calls outside opening hours - typically 40-60 calls per week that either go to voicemail or result in patients calling back, adding re-handling overhead.
After deploying an AI voice agent handling 60% of inbound calls autonomously:
| Metric | Before AI Voice Agent | After AI Voice Agent | Change |
|---|---|---|---|
| Weekly inbound calls | 320 | 320 (same volume) | - |
| AI-handled calls (60%) | 0 | 192 | +192 |
| Human-handled calls | 320 | 128 | -192 (-60%) |
| Missed calls (outside hours) | 50/week | 2/week (AI answers 24/7) | -48/week |
| DNA rate | 11% | 5% (AI reminder calls) | -6 percentage points |
| Receptionist hours on calls/day | 9.6 hours (2 FTE) | 3.8 hours (1 FTE) | -5.8 hours/day |
| Annual receptionist cost (calls) | £56,000 | £28,000 (1 FTE freed) | -£28,000 |
| AI voice agent annual cost | - | £4,200 (£0.11/min x 192 calls x 3min avg x 52wks) | +£4,200 |
| Net annual saving | - | - | £23,800 |
The one-time implementation cost of £5,000 (mid-range for this practice size and integration complexity) means the system reaches breakeven in approximately 2.5 months from go-live. The freed receptionist time is typically redeployed to patient-facing tasks - medication reviews, care navigation, patient registration - rather than being cut, which improves both patient experience and CQC ratings.
For dental practices, the ROI profile is slightly different. The higher average revenue per appointment (£85-£250 for private dentistry versus £26-£60 for NHS dental work) means that recovering even three or four missed appointment slots per week through after-hours AI booking has a disproportionate revenue impact. A private dental practice with 15 appointment slots per day and an AI voice agent recovering four previously-missed slots per week at an average fee of £150 generates an additional £28,800 in annual revenue - before any receptionist cost saving is counted.
Softomate Solutions offers a structured implementation package for AI voice agent deployment in UK medical and dental clinics. The package is designed to take a practice from initial enquiry to live deployment in 4-8 weeks, depending on the complexity of the clinical system integration and the number of call scenarios to be scripted.
The implementation follows six phases:
Phase 1: Discovery (Week 1) - A 90-minute structured discovery call with the practice manager and at least one clinician. Softomate maps the current call types, call volumes, telephony system, clinical management system, and key pain points. This session produces the Call Flow Specification document that governs all subsequent development.
Phase 2: Clinical Scenario Mapping (Weeks 1-2) - Every call scenario is documented in a structured format: trigger (what the patient says), AI response, verification steps required, API calls made, escalation conditions, and patient-facing confirmation. For a typical GP surgery, this produces 18-25 scenario maps. For a multi-site dental group, it can run to 40+. This document is reviewed and signed off by the practice manager before any development begins.
Phase 3: Integration Setup (Weeks 2-4) - Softomate configures the clinical system API credentials, sets up the telephony SIP integration, builds and tests the patient verification flow, and connects the booking and prescription routing functions. This phase includes the Information Governance agreement process with the clinical system vendor, which is often the longest dependency.
Phase 4: CQC Documentation (Week 3-4) - Softomate produces the written policies and procedures required for CQC compliance: AI Voice Agent Governance Policy, Escalation Protocol, Data Processing Record (Article 30 GDPR register entry), Staff Training Record template, and the Incident Response Procedure. These documents are reviewed with the Caldicott Guardian or Data Protection Lead before go-live.
Phase 5: Staff Training and UAT (Week 4-6) - A half-day training session with all reception and administrative staff. Covers: how the AI answers calls, when and how it escalates to them, how to access call transcripts, how to override the system, and what to do if the AI makes an error. User Acceptance Testing runs concurrently, with staff posing as patients to test all 18-25 scenarios.
Phase 6: Go-Live and Hypercare (Weeks 6-8 and 30 days post-live) - The AI goes live on a pilot basis, handling 20-30% of inbound calls initially, with human monitoring. Softomate provides 30-day hypercare: daily call transcript reviews, weekly performance reports, rapid iteration on any scenario that is not performing as expected. After 30 days, the system is typically expanded to full autonomous handling of the agreed call types.
Pricing ranges from £3,000 to £8,000 for the implementation package, depending on:
Ongoing monthly costs cover the AI platform usage (per-minute pricing, typically £300-£800/month for a single-site surgery at average call volumes) plus a £150-£250/month Softomate support and monitoring retainer. Total ongoing cost for a typical single-site GP surgery is £450-£1,050 per month - compared to the £2,300/month loaded cost of one full-time receptionist.
Yes, when correctly configured and documented. CQC does not prohibit AI voice agents - it requires that any technology used to handle patient interactions is safe, governed, and auditable under Regulation 17 (Good Governance). A compliant deployment includes a governance policy, staff training records, full call transcripts, escalation protocols, and UK-based data storage. Softomate provides all of these as part of the implementation package, producing the documentation that a CQC inspector would expect to review.
All patient call data - including transcripts and recordings - is stored in UK-based data centres by default in Softomate deployments. Patient call data is classified as special category health data under Article 9 of the UK GDPR. Softomate acts as a data processor under a formal Data Processing Agreement with each clinic. Data is encrypted at rest and in transit, retained only for the period specified in the practice's retention policy (typically 8 years in line with NHS record-keeping guidance), and never transferred outside the UK.
Emergency call handling is a mandatory configuration element in every Softomate healthcare deployment. When a caller uses any keyword associated with an emergency - chest pain, difficulty breathing, stroke, severe bleeding, unconscious - the AI immediately interrupts its normal flow, verbally signposts the caller to call 999, and simultaneously flags the call for urgent review by a staff member. The system does not attempt to triage symptoms. For urgent but non-emergency situations, callers are directed to NHS 111. This protocol is documented in the CQC governance policy.
Yes. Softomate has deployed AI voice agents in both NHS-contracted GP and dental practices and private clinics. NHS practices require additional attention to data governance (Information Governance agreements with EMIS or TPP, Caldicott Guardian sign-off) and access standards compliance. Private practices typically have a faster implementation timeline because the IG requirements are simpler. The core AI functionality - appointment booking, call routing, patient verification - is identical in both contexts.
Implementation costs £3,000-£8,000 as a one-time fee, depending on clinical system integration complexity and the number of call scenarios. Ongoing monthly costs are typically £450-£1,050, covering AI platform usage (per-minute) and Softomate's support retainer. For a surgery handling 300 calls per week with the AI managing 60%, the system typically reaches breakeven within 2-3 months through receptionist time saved and missed call recovery.
Yes. Multilingual support is a standard configuration option. Softomate's deployments on VAPI and ElevenLabs infrastructure support 29-32 languages, including Bengali, Gujarati, Punjabi, Urdu, Polish, Romanian, and Welsh. The AI can detect the caller's language preference from their opening words and switch automatically, or can be configured to offer a language menu at the start of the call. This is particularly valuable for practices in diverse urban areas such as Barking, Tower Hamlets, and Newham where a significant proportion of patients have limited English.
When the AI determines that a call requires human handling - because the patient is upset, the query is clinical, or the scenario falls outside the scripted flows - it performs a warm transfer. The receptionist receives the call with a brief spoken handover from the AI summarising the patient's name, verified identity, and the reason for transfer. A full transcript of the call to that point is simultaneously pushed to the receptionist's screen or sent as a notification. The patient does not hear hold music during the transfer - the handover is designed to feel seamless.
AI voice agents are a practical, CQC-compliant way for UK GP surgeries and dental clinics to manage the 60-70% of inbound calls that require no clinical judgement. With UK data residency on VAPI and ElevenLabs infrastructure, real-time integration into EMIS Web, SystmOne, and Dentally, and a structured 4-8 week implementation process that includes full CQC documentation, the technology is deployable today - not a future promise. A single-site GP surgery handling 300 calls per week can expect to recover £23,000-£28,000 in annual receptionist costs while eliminating missed calls and reducing DNA rates to below 6%.
Want to reduce your clinic's inbound call burden by 55-70%? Explore Softomate's AI Voice Agent for Healthcare service or book a free discovery call.
Written by Rakesh Patel, AI Automation Consultant at Softomate Solutions, Barking, East London. Specialising in AI voice agents and healthcare automation for UK clinics.More from the blog
Let us help
Talk to our London-based team about how we can build the AI software, automation, or bespoke development tailored to your needs.
Softomate Solutions Limited
Deen Dayal Yadav
Online
