Restaurant AI Ordering Assistant

Business Challenge

A restaurant chain was losing orders during peak hours. Their phone lines would light up from 11am to 2pm, and staff couldn’t keep up — calls went unanswered, customers got frustrated, and the chain was losing a significant portion of potential orders during lunch rush alone.

The problem wasn’t just volume. Phone orders were error-prone, averaging 3 minutes per call, and staff had no time to suggest add-ons or promotions. The chain had tried third-party delivery apps, but the 30% commission fees were eating into margins and they had no direct relationship with the customer.

Solution

We built a WhatsApp AI ordering assistant that lets customers browse the menu, customize their order, and pay — all through a conversation. The assistant understands natural language, so customers can say “I want a double cheeseburger with no pickles and extra cheese” and it parses the order correctly.

The system handles the full ordering lifecycle: menu questions (“Is the spicy sauce vegan?”), recommendations (“What’s good for a family of four?”), customization, upselling (“Would you like to add a drink for $2 more?”), and payment via WhatsApp Pay. Orders flow directly into the kitchen display system, and customers receive real-time status updates.

Technologies

The assistant uses the WhatsApp Business API for messaging, LangChain for conversation orchestration and natural language understanding, n8n for workflow automation connecting to the POS and kitchen systems, and PostgreSQL for storing customer profiles, order history, and menu data.

The menu is stored as a structured knowledge base that the agent queries, so price changes and new items are reflected instantly. n8n workflows handle the integration layer — syncing orders to the POS, triggering kitchen tickets, and sending delivery notifications — all without custom middleware code.

Results

Within three months of launch, a substantial share of all orders came through WhatsApp, completely transforming the chain’s order channel mix. The average order value increased meaningfully because the assistant consistently suggested add-ons and combos — something human staff forgot to do during rush hours.

Phone calls dropped significantly, freeing staff to focus on in-store customers and food preparation. Customer satisfaction improved because WhatsApp orders were faster, more accurate, and available 24/7. The chain also built a direct customer database of WhatsApp contacts that they now use for promotions and loyalty programs.

Screenshots

Sample WhatsApp ordering conversation with upsell

Order management dashboard showing WhatsApp order metrics

Architecture

The system is event-driven. When a customer messages the WhatsApp number, the WhatsApp API webhook triggers an n8n workflow that passes the message to the LangChain agent. The agent maintains conversation state in PostgreSQL, queries the menu knowledge base, and responds via the WhatsApp API.

Order completion triggers a separate n8n workflow that syncs to the POS, prints a kitchen ticket, and starts the delivery tracking loop. A session manager handles concurrent conversations, and a fallback handler routes complex cases to a human staff member who can take over the WhatsApp conversation from a dashboard.