How a boutique hotel automated 70% of its calls with AI (case study)
Starting point: a boutique hotel with an invisible problem
"Hotel Marqués de Atalaya" — name changed for confidentiality — is an 18-room boutique in Granada's old town. ADR €145, average occupancy 72%, 48% of revenue came from Booking and Expedia.
The problem: small reception (2 staff per shift), juggling in-person service, check-in and check-out, and incoming calls. 31% of calls went unanswered, and one out of every four answered ended up routed to Booking — costing 18% commission unnecessarily.
The GM put it bluntly: "We're paying commission on calls that are already coming to us. It's absurd."
What we tried before AI
- Part-time human staff for peak hours: cost €1,400/month, didn't cover nights.
- External call center: operators didn't know the hotel, gave wrong info, conversion dropped.
Neither solved the underlying issue: the hotel needed a voice channel that knew the product and was always on.
Implementation: 8 days
Day 1–2: discovery and content load
PDFs, Excel of room types, PMS data (Avirato), 25 recurring FAQs. Rinqa indexed it in under an hour.
Day 3–4: voice and scripting
Voice "Lucía" (Cartesia Sonic-2). Greeting: "Hotel Marqués de Atalaya, this is the assistant. How can I help you?". Transfer rules for explicit human request and group/event inquiries.
Day 5–6: PMS and email integration
Avirato API for real availability. Auto-email with Stripe payment link. Internal CRM webhook.
Day 7–8: tests and soft launch
50 simulated calls. Soft launch at 30% for 3 days. Prompt tuning. From day 9: 100% of traffic.
Results after 90 days
| Metric | Before | After | Change | |---|---|---|---| | Missed calls | 31% | 4% | −87% | | Resolution without humans | 0% | 71% | — | | Direct (non-OTA) bookings | 32% | 47% | +15 pp | | Avg call handling time | 4:12 | 2:38 | −37% | | Cost per booking captured | €26 | €4.20 | −84% | | NPS on calls | 7.1 | 8.4 | +1.3 | | Calls answered after hours | 0 | 187/month | — |
The big one: direct bookings rose 15 pp. Against monthly revenue of €78,000, that's €11,700/month no longer paid in OTA commissions.
What surprised them
- Automatic upselling (parking, breakfast, late check-out) increased average ticket 8%.
- Killer night-time conversions: 12% of bookings between 22:00 and 07:00 — high purchase intent.
- Front-desk briefing: automatic call summaries in the dashboard, no need to listen to recordings.
What didn't work at first
- The AI struggled with foreign names — fixed with Speechmatics ASR + vocabulary tuning.
- Confused "half board" with "all-inclusive" — fixed by sharpening the knowledge base.
- Didn't transfer corporate groups well — added explicit rule.
Does this work for every hotel?
Best fit: 12–80 rooms, >150 calls/month, defined product, PMS with API (Avirato, Mews, Cloudbeds, Mirai…). Less ideal: tiny hotels (<10 rooms — better with WhatsApp), large chains, rotating short-term rentals.
Real project cost
- Setup: €0
- Monthly fee: €199/month (up to 2,000 minutes)
- PMS integration: €0
- Year 1 total: €2,388
- Year 1 estimated savings: €140,400
ROI: ~58x.
Got a similar problem? Talk to the Rinqa demo for hotels and hear exactly how your virtual assistant would handle a real call.
