The Problem
Dimitri manages a busy Mediterranean restaurant in a tourist-heavy area. His WhatsApp number is printed on every menu, every table card, and on their Google Maps listing. Which means it's open season: questions come in from 7am until midnight, seven days a week.
When we first spoke, he walked us through what a typical day looked like. His front-of-house staff were fielding messages between seating guests: "Do you have vegetarian options?", "Can we book a table for 6 on Saturday?", "What time do you close?", "Is there parking nearby?". Each message takes 30–60 seconds to answer — but across 80+ messages a day with three staff sharing the phone, it added up to nearly 2 hours of lost floor time every single day.
He'd tried just leaving messages on read until quieter moments, but that was killing his Google ranking. Slow response rate on WhatsApp Business shows up in local search. He was stuck: respond fast and lose floor time, or protect the floor and lose visibility.
What He Actually Needed
After talking through the message types, a clear pattern emerged. The vast majority of incoming messages fell into four buckets:
- Reservation requests (date, time, party size)
- Menu and dietary questions (vegetarian, vegan, allergens)
- Basic info (hours, location, parking, dress code)
- Special requests (birthday setups, private hire)
Only the fourth category genuinely needed a human. The first three — easily 85% of all messages — could be handled automatically, instantly, and probably better than a rushed staff member typing on a phone between customers.
What We Built
The solution is a WhatsApp Business API bot powered by n8n, with Google Sheets as the reservation ledger and a simple static knowledge base for menu and FAQ responses. Here's the full flow:
The intent classifier reads the incoming message and routes it to one of four branches. Menu and info questions get an immediate templated reply pulled from a simple JSON knowledge base — allergen info, opening hours, location, dress code. Dimitri updates this file himself when the menu changes; it takes about two minutes.
Reservation requests kick off a short guided conversation: the bot asks for date, time, and party size, checks the Google Sheet for availability, and either confirms or offers alternative slots. The booking writes to the sheet automatically. Staff get a Telegram notification for every new reservation — no need to check the WhatsApp thread.
Anything the classifier isn't confident about — unusual requests, complaints, complex queries — gets flagged with a "human needed" tag and forwarded to the manager's personal WhatsApp with the full context. Response time on those drops from "whenever someone checks the shared phone" to instant.
The whole stack runs on n8n hosted on a $6/month VPS. WhatsApp Business API access was set up through Meta's Cloud API — no third-party subscription required, just a verified business account. Total build: 48 hours across two days.
The Edge Cases That Mattered
The core flow took about 6 hours. The remaining 42 were spent on the things that make or break a customer-facing bot:
- Handling ambiguous messages ("can we come in?" — bot asks for clarification politely)
- Same-day booking logic (different availability rules after 5pm)
- Party sizes over 10 — these auto-escalate to manager with a note about private hire
- Repeat messengers who've already confirmed — bot recognises the number and skips re-intro
- Out-of-hours responses — different message tone after midnight, sets expectation of morning reply for complex queries
None of these are technically hard. But skipping them is what turns a promising bot into an embarrassing one. We've seen restaurants try to DIY this with Zapier and a basic keyword filter, and the guest experience is rough. A bot that confidently misunderstands you is worse than no bot at all.
The Result
- 85% of messages now handled fully automatically — no staff time required
- Average first-response time: 2 seconds (was 28 minutes during service)
- WhatsApp Business response rate: now showing "Very responsive" on Google listing
- Staff reclaimed ~1.5 hours of floor time per day
- Reservation no-shows down 35% after adding automated reminder 2 hours before
- Zero monthly SaaS fees — just the VPS ($6/mo) and Meta's free API tier
The thing Dimitri keeps mentioning: guests actually comment on how fast they got a reply. "People assume there's someone on the phone constantly. They're pleasantly surprised — and that sets a good tone before they've even walked in."
That perception shift matters. A fast, accurate response to "do you have gluten-free options?" isn't just convenient — it signals that the restaurant is professional and attentive. The bot has become part of the brand experience.
What This Cost
This was scoped under our Growth plan — two connected flows (reservation + FAQ), WhatsApp API setup, custom knowledge base, and the Telegram escalation layer. Dimitri got full documentation, the n8n workflow JSON, and a 30-minute handover call. He's been running independently since week one.
If your restaurant, bar, or hospitality business is drowning in the same kind of messages, let's talk. The call is free and we'll tell you honestly what's worth automating and what isn't.