Pricing disclosure rules: what AI may quote vs. what must wait for human

The AI's biggest risk is mis-quoting price — it creates legal exposure (Thai consumer-protection rules), revenue leakage (under-quoting), and erosion of trust (over-quoting). This entry defines the boundary clearly.

What the AI MAY quote autonomously

These items are derived from structured data that is reliably available and unambiguous:

1. Published BAR (Best Available Rate) for a specific date range and room category, when:

• The PMS / rate system returns a single canonical number for that date+category
• VAT and service-charge inclusivity are recorded in the rate record
• The rate is not flagged as "promotional" or "do-not-quote-without-approval"

1. Standard published add-ons with fixed pricing:

• Airport transfer (one-way, by vehicle type)
• Standard breakfast (if not included in rate)
• Late checkout (if a published flat fee exists)
• Extra bed / rollaway (if published)

1. Policy-derived totals that are arithmetic:

• Multi-night totals at quoted nightly rate (3 nights × THB X = THB 3X)
• Two-room bookings at the same rate

1. Cancellation / deposit policy disclosure when stored as policy text — quote verbatim, do not paraphrase.

1. Tax breakdown when the inclusive total is quoted — "X total, of which 7% VAT, 10% service charge."

What the AI MUST escalate to human

Any of the following triggers escalation:

1. Custom packages — wedding inquiries, anniversary stays with add-ons, "we want a special arrangement"
2. Group bookings of 5+ rooms — even if the published rate technically returns; group rates are often negotiated below BAR
3. MICE / corporate RFPs — see mice-english-register for the full template; these never auto-quote
4. Rate parity questions — "Booking.com is showing X, can you match?" Always human.
5. Loyalty / repeat-guest discounts — these are relationship calls, not algorithmic
6. Long-stay rates (7+ nights, 14+ nights) — many properties have unpublished long-stay pricing
7. Negotiated discounts of any kind — "We're a large company, do you have a corporate rate?"
8. Anything during peak periods (Songkran, Christmas-NY, Chinese New Year) where the system returns a non-peak rate — likely data error; do not quote
9. Currency conversion to non-THB — quote in THB, let the human or the OTA handle conversion
10. Refund inquiries — policy-derived but emotionally charged; safer to route to human
11. Visa-related questions that intersect with stay duration — escalate; immigration is not a thing the AI should comment on

What the AI MUST disclose every time it quotes

• The dates the rate applies to
• The room category
• VAT and service-charge inclusivity (see thai-vat-conventions)
• Whether breakfast is included
• The cancellation policy (one line is sufficient: "Free cancellation up to 7 days before arrival")
• A clear next step: offer to pass the request to the reservations team to confirm the booking

What "escalate to human" means operationally

It does not mean "refuse to respond." It means:

• Send a holding reply: "Thank you for your inquiry. I'm passing this to our reservations team who will reply within [X] hours with a tailored response."
• Tag the conversation for the human queue with the reason (escalation: group-booking-5-rooms or escalation: wedding-package)
• Summarize the inquiry in the human-queue notes so the human doesn't re-read the full thread

The cost of getting this wrong

A wrongly quoted rate that the guest then books and shows up to honor is, in Thailand, often legally enforceable as the rate the hotel must accept (Thai consumer protection precedent leans toward the consumer on advertised price). The cost of a single mis-quote can be a several-thousand-baht concession. The cost of a correctly escalated inquiry is a 2-hour response delay. The math always favors escalation when in doubt.