Tools

What You'll Build

A WhatsApp agent that handles the repetitive front-desk messages and quietly passes the rest to a person.

A builder shared a working setup he made for a hospital. His flow, on WhatsApp, does four things: answers frequently asked questions by consulting a Google Sheet, handles appointment queries through an API, escalates to a human agent on request, and respects business hours so it only promises a human when one is actually available. It batches a customer's rapid-fire messages by waiting 20 seconds before replying, so it responds to the whole thought instead of half a sentence.

He posted it because it works but still trips up, and his stumbles are the most useful part of this playbook. You get his stack and his hard-won gotchas in one place.

Why This Works

Look at a day of messages to most small businesses and the pattern is obvious: "Are you open Sunday?" "How much is X?" "Can I move my appointment?" "Where are you?" The same questions, over and over, plus a steady trickle of booking requests. A person answers them between everything else, and the slow replies cost bookings.

WhatsApp is where those customers already are, which is the whole point. You are not asking anyone to call a number or use a web chat widget. The agent lives in the channel people already message you on, clears the repetitive 80 percent instantly, and routes the 20 percent that needs judgment to a human, with the context attached. The front desk stops being a bottleneck and the after-hours messages stop going unanswered until morning.

Prerequisites

Step-by-Step Setup

Step 1: Connect WhatsApp to n8n

Wire your WhatsApp number into n8n through Evolution API so incoming messages trigger the workflow and the agent can reply on the same number.

Step 2: Put Your FAQs in a Google Sheet

List the real questions and the approved answers, one per row. Keeping this in a sheet means your staff can update prices, hours, and policies without touching the workflow.

Step 3: Build the Agent and Its Instructions

The agent reads each message, checks the FAQ sheet for an answer, and decides whether this is a question it can handle, an appointment request, or something for a human. Keep the instructions focused (more on why below).

Step 4: Add the Appointment Path

For booking and reschedule requests, give the agent a tool: a call to your scheduling system's API, or a step that sends your booking link and confirms the details. This is the path that turns a question into a filled slot.

Step 5: Batch Fragmented Messages

People send three short messages in a row, not one tidy paragraph. Add a wait node (the source used 20 seconds) that gathers the customer's messages before the agent responds, so it answers the full request rather than the first fragment.

Step 6: Set Business Hours and a Human Handoff

Define when a human is available. During hours, "talk to a person" routes the conversation into Chatwoot for a staff member. Outside hours, the agent says so honestly and takes a message or books the callback, rather than promising a human who is asleep.

Step 7: Tune Against Real Messages

Run it on real conversations and watch where it fails. The first week is calibration, not launch (see Gotchas, this is where the build lives or dies).

Adapting This for Your Business

The hospital is incidental. Any business whose inbound is mostly questions plus appointments fits.

Gotchas and Tips

What This Replaces

Before this, fielding WhatsApp meant one of these:

After this, the repetitive questions and the simple bookings are handled the moment they arrive, day or night, the hard ones reach a human with context, and the cost is model usage plus tools you can self-host. You own the flow instead of renting it.

Keep Reading