Let me save you 20 minutes of reading vendor marketing: most AI customer service pitches for F&B are overselling. They show you a chatbot answering "What are your opening hours?" and call it a revolution.
The reality is more nuanced and more interesting. Here's what we've actually built and deployed.
The Real Problem in F&B Customer Service
If you run a food and beverage business — a restaurant chain, a café, a delivery kitchen — you already know the pattern:
Your staff gets the same 15-20 questions over and over. Menu items. Allergens. Opening hours. Delivery radius. Whether you accept large group bookings. Whether there's parking. Whether you do catering.
These questions arrive via Facebook Messenger, Zalo, Instagram DMs, WhatsApp, Google Business Messages, your website, and phone calls. Often simultaneously. During lunch rush.
Your staff answers these while also taking orders, managing the kitchen, and dealing with the customer standing in front of them. The choice is usually: ignore the messages or give hurried, incomplete answers. Neither option is great for business.
This is the problem worth solving. Not "revolutionising your customer experience with AI." Just: can we handle the repetitive queries automatically so your staff can focus on the humans in the room?
What We Actually Built
Our approach isn't a general-purpose chatbot. It's a purpose-built AI agent system — designed to do a specific set of things well, rather than everything poorly.
What It Handles (The 80%)
- Menu and product questions. "Do you have gluten-free options?" "What's in the pad thai?" "Can I see the drinks menu?" — answered instantly from a structured knowledge base that your staff maintains.
- Operating information. Hours, location, parking, delivery zones, accepted payment methods. The boring stuff that eats up 30% of your customer messages.
- Reservation and booking inquiries. Captures party size, preferred date/time, contact details, and special requests. Doesn't replace your reservation system — feeds information into it.
- Order status updates. When integrated with your delivery or POS system, the AI can pull real-time order status without anyone in your team touching it.
- FAQ routing. Allergen policies, refund windows, loyalty programme details — anything that has a definitive answer gets answered.
What It Doesn't Handle (The 20%)
- Genuine complaints. If someone received the wrong order or had a bad experience, the system escalates immediately to a human. AI should never attempt to resolve emotional customer complaints — that requires empathy, not patterns.
- Complex custom orders. "Can you do a modified version of the set menu for 40 people with 3 vegetarians, 2 halal, and one person who's allergic to everything?" — that goes to your events coordinator.
- Anything ambiguous. If the AI isn't confident in the answer, it admits it and routes to a human with full conversation context. No hallucinating menu items.
The 80/20 split isn't a marketing claim. It's based on message analysis across our deployments. The vast majority of customer queries in F&B are repetitive and factual — exactly what AI handles well.
Why You Don't Need an ERP or CRM to Start
This is the part most vendors get wrong: they tell you to buy their CRM platform first, then add AI on top. That's backwards.
Here's what you actually need:
- A structured document with your menu, prices, hours, policies, and FAQs. A Google Doc works. A Notion page works. A spreadsheet works. The AI needs a knowledge source — it doesn't need to be fancy.
- Access to your messaging channels. Facebook Messenger API, Zalo OA, website chat widget — wherever your customers actually reach you.
- A human escalation path. Someone who will respond when the AI routes a conversation that needs a real person. This can be the same staff member who was handling all messages before — they'll just have 80% fewer to deal with.
That's it. No ERP. No CRM. No six-month integration project.
If you do have an ERP or POS system, we can integrate with it for order status and inventory queries. But it's not a prerequisite. Start simple, expand later.
The Data Integration Layer
For businesses that want to go further — connecting their POS, inventory, and delivery systems so the AI has real-time data — we use Node-RED as the integration backbone.
Node-RED's advantage for F&B specifically:
- It connects to anything. Legacy POS systems with bizarre APIs? Zalo's webhook format? A supplier's Excel-based ordering system? Node-RED has nodes for all of it.
- Visual flow design. Your operations manager can understand the data flow without reading code. This matters when the person maintaining the system isn't a developer.
- Runs on minimal hardware. We've deployed Node-RED instances on Raspberry Pis sitting in restaurant back offices. Monthly hosting cost: the electricity bill for the Pi.
A typical integration flow: POS transaction → Node-RED processes the event → updates a lightweight database → AI agent queries the database when a customer asks "Where's my order?"
No data warehouse required for this level. If you need enterprise analytics across 200+ outlets, that's a different conversation.
What It Costs
I'll be straightforward because nobody else in this space is: the total setup cost is roughly comparable to one or two months of a staff member's salary. The ongoing cost is a fraction of that.
For context — hiring someone in Ho Chi Minh City to handle messages full-time runs $400-600/month in salary alone, plus training time, turnover, and the reality that humans need sleep and holidays. An AI system that handles 80% of those messages costs significantly less and runs around the clock.
The economics are simple. The harder question is whether the system actually works well enough to trust — which is why the next section matters.
What We Learned the Hard Way
A few things we got wrong in early deployments that you should know about:
Tone matters more than accuracy. An AI that gives the right answer in a robotic tone loses customers faster than a human who's slightly late but warm. We spent significant time tuning response personality — our agents are friendly and concise, not corporate.
Knowledge bases go stale. Your menu changes. Seasonal items come and go. If nobody updates the AI's knowledge source, it starts confidently telling customers about dishes you stopped serving three months ago. We now build update reminders into the workflow.
Multilingual is non-negotiable in Vietnam. Customers switch between Vietnamese and English, sometimes mid-sentence. The AI needs to handle both fluently, including Vietnamese slang and regional variations. This was harder than expected.
The Honest Assessment
AI customer service for F&B isn't going to transform your business overnight. It's not a revenue driver — it's a cost reducer and a quality-of-experience improver.
What it does: frees your staff from the message treadmill so they can focus on the customers physically present. Ensures no inquiry goes unanswered at 2 AM. Provides consistent, accurate information regardless of which shift is working.
What it doesn't do: replace the human touch that makes hospitality special. Cook better food. Fix a bad location.
If your F&B business is drowning in repetitive customer messages and your staff can't keep up, this is the specific problem these systems solve well. We've written more about the architecture behind our AI agents and the persona engineering that makes them actually useful.