Getting AI live at a boutique hotel is a two-week project if you do it in the right order: get the data foundation clean, connect WhatsApp, build the knowledge base, set guardrails, soft-launch with staff shadowing, then go live and tune. The mistake that wastes months is starting with the AI before the underlying PMS data, payments, and policies are in order — the agent is only as good as what sits beneath it.

Use this checklist top to bottom.

1. Foundation — before any AI

Your PMS has accurate room categories, occupancy, and bed configurations.
Rates are correct and date-based (including weekend and season), with GST slabs set.
Policies are defined: check-in/out, cancellation, child and extra-bed, advance payment.
A payment gateway (Razorpay/UPI) is connected and tested.
Your OTAs are synced through a channel manager so availability is truthful.

The agent quotes and books from this data. If it is wrong here, the AI will be confidently wrong to guests.

2. Connect the WhatsApp Business API

Provision a dedicated number not already on WhatsApp.
Verify the hotel as a business in Meta Business Manager.
Get core templates approved in the correct category (confirmation and pre-arrival as UTILITY; review requests as MARKETING).

Full walkthrough: how to set up the WhatsApp Business API for a hotel.

3. Build the knowledge base

Every room category and what makes it distinct.
Rate plans and meal plans, with inclusions.
All policies, amenities, and accessibility details.
Local context: directions, airport distance, nearby attractions.

This is the single biggest lever on answer quality, as covered in how to train an AI agent for hotel reservations.

4. Set policy guardrails

Advance-payment rule encoded so the agent never promises "pay at hotel" against policy.
GST and pricing rules so quotes are always correct.
A no-hallucination rule — never claim availability or amenities the data does not support.
Post-response checks that rewrite any message drifting past policy.

5. Define escalation

Complaints, refunds, and disputes route to a human.
Large or complex group bookings route to a human.
Anything outside the knowledge base escalates with full context.

6. Soft-launch with shadowing

Enable for returning guests or one channel first.
Have staff review every response for the first ~100 conversations.
Feed every correction back into the knowledge base and settings.

7. Go live and promote

Open to all guests once shadowing looks clean.
Publish the WhatsApp number on your website, Google Business Profile, social, and OTA messaging.
Make it as prominent as your phone number.

8. Tune continuously

Review transcripts weekly.
Track resolution rate, inquiry-to-booking conversion, first-response time, and any hallucination.
Refine the knowledge base and prompts; the agent should improve monthly.

How Hotelary shortens the checklist

Because Hotelary's WhatsApp AI runs on the same system as the PMS, steps 1, 3, and 4 are largely done by your normal hotel setup — the agent inherits live rooms, rates, policies, and GST as its knowledge and guardrails. Hotelary also manages the WhatsApp connection and template provisioning, so a boutique property can reach soft-launch in days rather than weeks. See the week-by-week version in our AI concierge implementation guide.