AI Booking Automation for a Dental Practice

1. The problem

A multi-chair Cambridgeshire dental practice — NHS + private mix — was drowning reception in routine bookings, reminders, and recalls. Reception spent 60-70% of their day on phone bookings. Patients increasingly expected to book online (they do everything else online), but the practice management system (Dentally) had a basic booking widget that didn't handle the practice's capacity rules. No-show rate was ~11%. Recall outreach was inconsistent — done when reception had time, often months late. After-hours email enquiries went unanswered until Monday.

2. Stack

3. Architecture

System topology — what runs where, what talks to what:

        +-------------------+
        |  Patient on phone |
        |  or computer      |
        +---------+---------+
                  |
                  v
        +-------------------+         +----------------------+
        |  Next.js booking  |         |  Vercel serverless   |
        |  app              | ------> |  function            |
        |  Mobile-first     |         |  - Verify identity   |
        +-------------------+         +----------+-----------+
                                                 |
                                +----------------+----------------+
                                |                                 |
                                v                                 v
                +-------------------------+      +---------------------------+
                |  Dentally API           |      |  Local booking cache      |
                |  - Live slot avail.     | <--> |  (Postgres, 5-min TTL)    |
                |  - Book/reschedule      |      |  Avoids hammering Dentally|
                |  - Patient lookup       |      +---------------------------+
                +-------------------------+
                                |
                                v
                +-------------------------+      +---------------------------+
                |  SMS reminders          |      |  Recall workflow          |
                |  (Twilio)               |      |  6-month check-up due ->  |
                |  - 24h + 2h before      |      |  auto SMS w/ booking link |
                |  - Reschedule link      |      |  - tracked in Postgres    |
                +-------------------------+      +---------------------------+

        +------------------------------------------------------------------+
        |  After-hours triage chatbot (GPT-4o-mini, grounded in FAQ)       |
        |  - Books routine appointments online                              |
        |  - Triages emergencies (toothache, swelling) -> emergency contact |
        |  - Answers admin Qs (hours, prices, services, what to bring)      |
        |  - Auto-escalates uncertain cases to "we'll call you back" queue  |
        +------------------------------------------------------------------+

4. Automation flow

End-to-end runtime flow — what happens when a real input arrives:

1. Routine booking. Patient visits practice website, picks service + practitioner + time. Real-time slot availability via Dentally API. Confirmed in 30 seconds. SMS + email confirmation sent.
2. Reminders. Lambda runs hourly. For each appointment <24h away, send SMS reminder with reschedule link. Same again 2h before. Confirmation request: "Reply Y to confirm".
3. No-show prevention. Patients who don't confirm 2h before get a follow-up call from reception (now only ~3% of bookings vs ~30% pre-automation). Confirmed-but-no-show rate drops from 11% to ~3%.
4. Recall workflow. Daily Lambda checks Dentally for patients whose last check-up was 5.5 months ago. Auto-SMS sent with one-tap booking link. Tracked in Postgres for follow-up at 6, 7, 9 months if no engagement.
5. After-hours triage. Chatbot embedded on website. Available 24/7. For routine queries → books or answers. For potential emergencies → routes to emergency dental line + creates ticket for morning reception review. Uncertain cases → "we'll call you back" queue.
6. Audit trail. Every booking, reminder sent, recall outreach, chatbot interaction logged to Postgres. Compliance-grade retention.

5. What success looked like

• Reduce reception phone time on routine bookings by 60%+
• Cut no-show rate from ~11% to under 5%
• Restore consistent recall outreach (was sporadic)
• Provide after-hours response without staffing it
• No disruption to existing Dentally workflow — practice continues using their familiar system

6. Outcomes

7. Speed improvements

• Booking time. Phone booking: avg 4 minutes (waiting on hold + reception time). Online booking: ~30 seconds.
• Reception availability. Reduced inbound routine calls = more reception time for higher-value calls (complex private consultations, emergency triage).
• After-hours response. Was: nothing until next morning. Now: instant (chatbot) or "we'll call you back" queue created.
• Recall outreach latency. Was: weeks/months late. Now: same-day on the 5.5-month anniversary.

SEO & visibility growth

• Google reviews. +19 reviews in first 3 months (vs ~2/month historical baseline) — driven by automated post-appointment review request
• GBP impressions. +41% (higher review count + faster GBP response time improves local ranking)
• "dental practice [town]" rank. Moved from page 2 to top of page 1 in primary catchment over 6 months

9. ROI math

10. Maintenance model

See care plan tiers for full structure.

11. Honest gotchas — what we'd do differently

• Dentally's API has appointment-type complexity (NHS band 1 vs band 2 vs band 3 vs private vs hygienist) — needed careful schema mapping. Documented thoroughly because the practice may want to extend later.
• After-hours emergency triage requires extreme care — chatbot has hard rules for "swelling", "trauma", "facial swelling" → immediate escalation to emergency dental line, never tries to book. Got this wrong in week 2 (system booked an emergency for a Tuesday), retuned with explicit hard-block keywords.
• GDPR / clinical data: implemented data minimisation aggressively. Patient identifier (NHS number) only used for booking, not stored in long-term cache. Conversation transcripts retained 30 days.
• SMS deliverability for older patients on PAYG carriers was 88% (vs 99% on contracts) — added email fallback for confirmed no-deliveries.
• Self-serve adoption was slower than expected for under-30s (they'd still phone — possibly because dental anxiety drives toward voice contact). Adoption only hit 58% after we added a clear visual difference between "for a routine check-up" (online) and "for something urgent" (phone).
• Recall workflow needed practice-manager approval mechanism — auto-sending recall SMS to bereaved family members was a real concern. Built in a "do not contact" tag with full-team training.