How I’m Prototyping an AI Agent Reservation System for a Filipino Private Dining Restaurant Using ManyChat + n8n

Hi again, this is G — and this weekend, I built something simple but potentially powerful: an AI-assisted reservation system for a Filipino-owned private dining spot.

It’s not a polished product yet, but it’s a working MVP that I’ll be showing to the client this week. Bonus? It also doubled as my re-introduction to ManyChat, which I hadn’t touched seriously in over a year. And honestly, I’m kinda impressed by how much the platform has improved and how seamlessly it integrated with n8n for the webhook automation.

This project wasn’t just about building a tool — it was a way to help a local restaurant modernize their booking process while allowing me to test-drive a lean, low-code setup that could be reused for other Filipino businesses too.

What I Wanted to Build

The client — a boutique private dining restaurant in the Philippines — handles most of their inquiries and bookings via Instagram DMs. They don’t use booking forms or apps. Everything is managed through chat — manual, repetitive, and often prone to human delay.

So the idea was this: Can we create an automated reservation assistant using Instagram + ManyChat that still feels conversational but handles the repetitive parts?

The goal wasn’t to replace the human connection but to simplify it — collect essential booking info upfront so the owner doesn’t have to keep asking the same questions every day.

The basic booking flow needed to capture:

• Name
• Date of reservation
• Contact number
• Optional special requests (like allergies or seating preferences)

All this would then be logged into:

• Google Sheets – for easy viewing and manual edits
• Google Calendar – to visualize confirmed bookings and reduce double-booking risks

This wasn’t just about the tech — it was about translating hospitality’s warm touch into a workflow that respected the client’s existing tools and behaviors.

Tech Stack (Simple but Effective)

Here’s what I used:

• ManyChat – to create the Instagram message flow and collect structured responses from users. (Think: "Hi! What day would you like to dine with us?")
• n8n – my favorite open-source automation tool, to receive data via webhook and connect it to other tools.
• Google Sheets – to log every confirmed reservation, which the restaurant owner is already familiar with.
• Google Calendar – planned as the visual layer to display upcoming bookings and time blocks.

All of these were tied together using a simple webhook from ManyChat → n8n.

And it worked seamlessly.

Why ManyChat (Again)?

I’ve tinkered with ManyChat in the past — but back then, it felt overwhelming, especially for non-Facebook channels. This time, using it with Instagram + webhooks + my own stack? It finally made sense.

What I liked:

• Setting up IG-flows was surprisingly intuitive
• Webhook integration with n8n was fast and painless
• Conversation design felt light and warm, not robotic
• It worked inside Instagram, where the client already lives (this is huge)

What I wish:

• Only business connection with TikTok (although it is still in beta)

Still — for a first-pass MVP? Sulit. And more than enough for this type of business.

How It Works (AI DM Bot Setup with ManyChat + n8n)

This setup is inspired by a build I saw from Yashika of Automates AI. It’s a low-code Instagram DM bot that captures leads, qualifies them through natural conversation, and saves info into Google Sheets — all without going through Facebook’s Meta Business verification.

Let me walk you through it.

1. ManyChat Trigger Setup

• In ManyChat, create a new Automation → Start from Scratch
• Choose trigger: Instagram > User sends a message
• Select: "Detect Specific Words in a Message" (e.g., checklist, info, reserve)

2. Define User Fields

• Go to Settings > Fields > User Fields
• Create user_input and assistant_reply as custom fields to store incoming messages and outgoing AI replies

3. First Message Response

• Set Action: Set User Field → user_input = last text input
• This is the value you’ll send to n8n

4. Setup n8n Webhook Endpoint

• In n8n, create a new workflow
• Add node: Webhook
  • Method: POST
  • Respond via: Webhook Response Node
• Copy the webhook URL

5. Back in ManyChat → Send Webhook

• Action: Make External Request
• Method: POST
• URL: Your copied webhook URL
• Body: Send JSON like:

{
  "user_input": "{{user_input}}",
  "contact_id": "{{contact.id}}"
}

6. Add AI Agent Node in n8n

• Node: AI Agent
• Prompt: Use user_input as the user message
• System prompt example:

You are SUPER SARAP Restaurant’s friendly Instagram DM assistant. You help pre-qualify leads for their private dining experience by asking questions, capturing name and email, and only sending a booking link to qualified leads.

7. Memory

• Enable memory by setting key to contact_id
• Retain last 10 messages for context

8. Parse the AI Output and Respond Back

• Add: Respond to Webhook
• Add: Structured Output Parser if using a strict JSON format

9. Back in ManyChat: Receive Response

• Add: Response Mapping
  • JSON path: $.0.output
  • Target Field: assistant_reply

10. Respond via Instagram Message

• Message Block: use {{assistant_reply}} as response
• Assign next user input to user_input again (loop back)

11. Then Store in Google Sheets

• In n8n, add node: Google Sheets > Append Row
• Capture name, email, and summary extracted by AI
• Match this to the same variable names in the AI response

This creates a powerful conversational funnel that runs entirely inside IG DMs, stores your leads in Sheets, and does real-time qualification.

If you want to customize the flow:

• Change prompt questions
• Adjust trigger keywords
• Use CRM integrations

Learnings from This Build

This build reminded me why I love scrappy, real-world MVPs.

• Simple chat-based flows can handle 80% of what service-based businesses need. These kinds of conversational interfaces allow you to qualify leads, collect data, and even trigger automations without building a full-blown app.
• ManyChat + n8n is a dream combo for fast, repeatable builds. It’s modular, scalable, and allows you to inject AI into conversations without extra platforms.
• This is doable even if you’re not a coder — it’s all config and logic. You don’t need to write JavaScript or wrangle APIs manually. Just map your variables, write clear prompts, and you’re good.
• The best testbed is your own IG account — you feel what the user feels. Before showing it to the client, I tested every flow inside my personal IG DMs. It helped me catch awkward moments and refine my message blocks.

Also: I found myself comparing this setup to Voiceflow. While Voiceflow is fantastic for building voice agents and complex multi-turn logic trees (especially for smart speaker flows or advanced decision trees), ManyChat wins on speed, simplicity, and deployability — especially when your audience already lives on Instagram and expects fast, personalized replies.

In short: If your goal is to convert DMs into leads or bookings today, ManyChat + n8n is the shortest path to value.

Gotchas and Tips

• Webhook testing: Always send test payloads from ManyChat with realistic values. Helps with mapping.
• Phone number validation: Add a confirmation step. I added a fallback like: "Is this the right number to confirm your booking later?"
• Timezone issues: If you’re logging to Google Calendar, double-check timezone settings or you’ll end up with mismatched times.
• Preview mode: Test flows inside your own IG to see how it feels. Better than using simulator.

What’s Next?

This coming week, I’ll be showing the flow to the restaurant owner. I’m keeping it flexible so they can request tweaks. But the foundation is ready.

Here’s what’s on my improvement list:

• Finalize Calendar sync with Google Calendar (via n8n) so they can see bookings at a glance
• Experiment with GPT fallback replies via OpenAI node in n8n for weird user responses
• Set up a daily digest email of bookings using Gmail node
• Long-term? Could extend this to other channels like WhatsApp or FB Messenger

And who knows — if this works well, it could be the seed of a broader booking CRM for Filipino-owned food businesses. Something light, chat-first, and easy to use.

🌱 Closing Thoughts

Sometimes, all you need to validate an idea is a working flow.

It doesn’t have to be perfect. It just needs to work. It just needs to answer the need.

If you’re a builder looking for a weekend project or a service provider who wants to test a lightweight automation — start with chat.

Instagram + ManyChat + n8n is a powerful and underused combo.

Don’t wait for a full design sprint. Try something. Show it. Adjust. That’s what this was for me — and I’m proud of how fast it came together.

Want to see the flow? DM me.
Built something similar? Let’s swap notes.

— G