1. Snapshot: where medical-aesthetics clinics stand on AI
Aesthetics is a high-ticket, high-inquiry, high-repeat business — which means every missed LINE message and every no-show is money walking out the door. Taiwan's medical-aesthetics market crossed NT$60 billion in 2024, up more than 40% in five years, with roughly 445 licensed clinics nationwide — nearly 165% growth in five and a half years, so competition is intensifying fast. The global market is expanding at about 11.9% CAGR. Yet most clinics are still stuck on "one LINE account answered by hand, one Excel sheet for scheduling." Evening and weekend inquiries go dark — exactly the gap AI fills most easily.
2. Three to five typical AI use cases
Not every scenario is worth doing on day one. Ordered by "sharpest pain, fastest cash":
- LINE / AI consult auto-reply: turn price bands, aftercare, hours, and parking into a knowledge base that answers instantly after hours, plugging the "seen-but-ignored" leak.
- Appointment and no-show reminders: automatic 24-hour and 2-hour reminders with one-tap reschedule. The global healthcare no-show average is about 23.5%, and systematic reminders cut no-shows by 30-60%.
- Treatment recommendation and segmented remarketing: auto-segment by last treatment and interval (e.g., HA fillers at 6 months, laser off-season win-back) and push personalized return messages.
- Post-treatment follow-up and education: automated D+1 / D+7 / D+30 check-ins that gather recovery status and prompt review visits, lowering complaints.
- Google reviews and reputation management: guide happy clients to review, alert staff instantly to negatives, and draft tactful replies with AI.
3. Two real cases
Clinic A (Greater Taipei, single site, 3 rooms): before AI, about 90 LINE inquiries a day handled by two front-desk staff, evening messages left until the next day, an estimated one-third lost. After deploying an AI agent plus knowledge base, about 70% of common questions (price, aftercare, hours) were auto-answered, freeing staff for complex cases. Three months later, LINE inquiry-to-booking conversion rose from 18% to 26%. The pitfall: early on the knowledge base didn't lock down "no efficacy claims without a doctor's diagnosis," and the bot nearly promised results — later rewritten to "always route to an in-person consult, never discuss individual efficacy."
Peer B (chain, 4 branches): focused on no-shows and remarketing. After integrating the booking system, two-stage 24h/2h reminders plus one-tap reschedule cut no-shows from 21% to 9%; segmented win-back during laser off-season brought back 40-plus lapsed clients in a single month. Biggest lesson: initial push frequency was too high and got them blocked — capped at "2 messages per person per month," opt-outs stabilized.
4. Recommended tool stack
In aesthetics the core is "LINE as the main battlefield, plus the ability to hook into bookings and client data." Suggested build:
- Conversation and knowledge base: OpenAI API or Claude for language understanding, with RAG reading the clinic's own FAQ so the model can't invent efficacy claims.
- Workflow automation: n8n or Make to connect the LINE Messaging API, booking system, and Google Sheets / CRM for reminders, segmentation, and pushes.
- Front-end entry point: LINE Official Account (LINE OA) primary, a website chat widget secondary.
Why n8n: self-hostable, keeps data in your own hands (aesthetics involves personal and health data with high leakage risk), and monthly cost stays more controllable than black-box SaaS over time.
5. ROI model
For a single site with 2-3 rooms and ~1,500-2,500 monthly inquiries:
- Investment: one-time build (LINE agent + knowledge base + reminders + segmentation) roughly NT$120,000-250,000; monthly run cost (API usage + hosting + ops) roughly NT$4,000-9,000.
- Output (labor saved): automating 70% of common inquiries equals ~0.5-1 front-desk headcount, saving ~NT$15,000-30,000/month.
- Output (revenue added): cutting no-shows from 21% to 9% — at 300 bookings/month and NT$3,000 average contribution — recovers ~36 visits/month ≈ NT$108,000; remarketing win-back adds tens of thousands more.
- Payback: for the agent + reminders combo, most clinics land at 4-8 months; adding segmented remarketing contributes faster but needs clean client data.
6. Rollout timeline (Phase 1-4)
- Phase 1 (weeks 1-2) assess and inventory: organize existing FAQ, booking flow, and client list status; define compliance red lines.
- Phase 2 (weeks 3-5) agent and knowledge base go live: start with after-hours auto-reply, staff can take over anytime.
- Phase 3 (weeks 6-8) reminders and no-show: integrate the booking system; enable 24h/2h reminders and one-tap reschedule.
- Phase 4 (weeks 9-12) segmentation and remarketing: only after data is clean, run segmented pushes with a frequency cap and opt-out.
7. Common failure modes and how to avoid them
- Loose knowledge base making efficacy promises: explicitly forbid the AI from discussing individual efficacy, always "route to consult," and human-review high-risk answers.
- Over-frequent pushes getting blocked: set a per-person monthly cap, segment sends, keep one-tap opt-out.
- Dirty client data: dedupe and re-tag before segmenting, or win-back messages hit the wrong people and cost trust.
- Treating AI as fully autonomous, unsupervised: medical context needs human review; route to a human when confidence is low to avoid awkward or compliance incidents.
8. Scenarios NOT suited to AI
- Individual indications and risk assessment needing a doctor's judgment — AI can only route to a consult, never decide.
- Post-treatment anomalies or suspected complications — must escalate to a human and doctor immediately.
- Disputes, refunds, complaint escalations — emotional and legal; AI only triages and notifies.
- Marketing copy involving efficacy, comparison, or guarantee wording — needs human plus compliance review.
9. How ScriptWalker fits
ScriptWalker is a Laravel + Flutter studio. We wire your LINE OA, booking system, client data, and AI agent into one maintainable flow, with data staying on your own server. Adoption starts from NT$80,000 (depending on systems integrated and knowledge-base scope); you can start with the minimum viable "agent + reminders," see the numbers, then expand to segmented remarketing. On medical-advertising compliance we flag the common red lines, but ScriptWalker is not a legal advisor — final ad copy should still be confirmed by the clinic and a qualified compliance officer or lawyer.
10. FAQ
Will the AI agent go off-script and make illegal efficacy promises?
That depends on design. Using RAG, we let the AI read only the clinic's approved knowledge base, explicitly forbid discussing individual efficacy, always route to an in-person consult, and hand off to a human when confidence is low — keeping compliance risk minimal.
How long does it take? Is it a hassle?
The minimum viable version (agent + reminders) usually goes live in 4-6 weeks. You just provide your existing FAQ and booking flow; we handle the integration.
Roughly how fast is payback?
For a single site, labor saved plus lower no-shows put most clinics at 4-8 months; adding remarketing is usually faster but needs clean client data.
Is client personal and health data safe?
We recommend a self-hosted flow (e.g., n8n) with data kept on your own server, not handed to black-box SaaS, and minimal collection plus access control per data-protection law.
Can I do just one part?
Yes, and we recommend it. Start with the fastest-payback "agent + reminders," then expand to segmentation and remarketing once you see the numbers.
11. Decision checklist + call to action
- ☐ Do evening/weekend LINE inquiries at my clinic often wait until the next day?
- ☐ Is my no-show rate above 15%?
- ☐ Do I have a client list but almost never run segmented win-back?
- ☐ Do common questions (price, aftercare, hours) eat a lot of front-desk time?
- ☐ Is post-treatment follow-up done by hand and often missed?
- ☐ Do I often notice Google negatives too late?
- ☐ Is client personal/health data scattered in Excel with weak control?
- ☐ Am I clear on medical-advertising red lines (no individual efficacy, no guarantees)?
- ☐ Do I have budget for a minimum viable "agent + reminders" first?
- ☐ Do I want data kept on my own server rather than in black-box SaaS?
Tick 3 or more and it's worth building a minimum viable aesthetics AI flow first. Book a consult:
- Email：[email protected]
- 電話：0916-224-047
- LINE：@ufv9089p