You get a WhatsApp message at 3am. "Do you have parking?" It's a simple question. Your night reception is either asleep or non-existent. Your team sees it at 9am. By then, that customer has already booked with the hotel down the road.
This is the problem "automatic WhatsApp answers" is supposed to solve. But most of the solutions on the market are scripted chatbots — flow diagrams pretending to be conversation. Below: why those fail, what real automatic answers require, and how to set it up in a day.
Why scripted chatbots fail
A scripted bot works like this: "If the user says X, reply Y." You sit down, list every question your customers might ask, and write a response for each. The problem: your customers don't phrase questions the way you wrote them. A hotel bot with "do you have parking?" as a trigger won't match "is there anywhere to leave the car?" — and the customer walks.
The fix isn't more scripts. The fix is an AI that reads your actual documents (menu, FAQ, room list, policies) and answers questions in the customer's own words, in the customer's own language, even ones you never anticipated.
What real automatic answers need
It reads your real content
Not a bot flow. Not a decision tree. An AI that reads your PDF FAQ, your Google Doc menu, your CSV price list. Ask "how much is a family suite for 3 nights?" — it does the math from your pricing table.
It handles rephrasings
A customer might ask "are you dog-friendly?" or "can I bring my labrador?" or "no pets allowed, right?" — three phrasings of the same question. Real AI reads intent, not keywords. Scripted bots need three separate rules.
It speaks the customer's language
Auto-detect + reply in the same language. A customer writes in Portuguese, gets a Portuguese answer. Same knowledge base, no translation layer, no per-language rebuild. Aim for 40+ languages out of the box.
It escalates when unsure
The AI should default to "I don't know — let me get someone" rather than hallucinate an answer. Every escalation lands with your on-duty operator via app notification. The customer sees a human take over mid-conversation.
How to set it up in a day
- Hour 1. Send your existing docs — menu, FAQ, price list, schedule, policies. Google Docs, PDFs, Word, plain text, or a public URL. The AI indexes them.
- Hour 2. Connect WhatsApp Business. You use your existing number or we help you get a new one. (Meta's side takes 3-5 days for a new WABA; existing numbers work day-of.)
- Hour 3. Start in "assisted mode" — the AI drafts every reply, but doesn't send until your team clicks approve. You watch what it drafts, tweak the KB where needed.
- Hour 4. Flip to auto for one channel and one topic (say, "menu + hours"). Monitor. Expand as trust grows.
- End of day. Your team is answering 70-80% fewer messages, all after-hours messages get answered, and the customer never knows the difference.
Assisted mode: the safety net
New to AI? Start in assisted mode. Every AI-drafted reply lands in a review queue on your operator app. Tap Approve to ship it, Edit to tweak, or Skip to reply yourself. When you trust the answers on a topic, one click flips that topic to auto. The AI learns your tone from every edit — so week 3 needs less approval than week 1.
FAQ
What if the AI makes something up?
It doesn't. It answers only from the KB you provided, plus the current conversation. If it doesn't know, it says so and routes to a human. No hallucinated tee times, no invented prices.
What about voice notes?
Transcribed, understood, answered. If you want, the AI can reply with voice too (text-to-speech).
What about photos?
Vision-capable AI: menus, receipts, screenshots, confusing signs. Reads them and replies in words the customer can act on.
Do I lose my team's judgement?
No. Auto / Assisted / Manual — you pick how much the AI handles, per topic, per instance. Your team stays in charge of what matters.
How long until it's trained?
It's "trained" the moment you upload docs. It gets better with every operator edit, but usable from message one.