A WhatsApp agent that sends on its own will eventually quote the wrong rate to the wrong guest at the worst hour. The draft plus approval pattern keeps the speed and drops the risk.
Three steps
- Load the context before the words. The agent reads the thread, pulls the reservation and whatever was already promised, then writes. An agent with no booking record in front of it invents one.
- Route the draft, not the reply. The output lands where a human already works: a CRM task, a shared label. Nothing reaches the guest until a person presses send.
- Hard-code two lists. Name what the agent may draft for approval, and what it hands straight to a person without drafting anything.
Two levels
Draft, then approve. Rate quotes, discounts, upgrades, extra nights and cancellations within policy. The agent pulls the rate from your PMS or RMS instead of writing it, and a person checks it before it goes out. This is where the revenue is: an extra night to fill a gap or an early check-in is money you keep, so these stay in the draft queue, not on a never list.
Straight to a person, nothing drafted. Complaints, injuries, data requests and refunds outside policy. Those carry your name, not the model's.
What happens at night
An approval gate means nobody approves at eleven at night. Plan for it: the only thing that goes out on its own is a pre-written acknowledgement that the message arrived and when a person will reply. The guest knows within seconds that someone has the message, and gets the checked reply when the team is back.
Rafael Osborne (Profit Guest Services): start with pre-stay and arrival-day questions, the highest volume and the lowest risk. Same filter as in what not to automate in a small hotel.
Does approval kill the time saving? No. Drafting is the slow part. Approving a FAQ answer takes seconds. A rate quote still needs checking against your system. That is what the gate is for.
