Industrial distribution · Sales and operations
AI Sales Agent
An AI agent that answers WhatsApp, qualifies every customer and hands ready-to-do work to Sales, the Workshop and Management.
- Client
- Industrial safety equipment distributor
- Role
- Architecture, automation and full-stack development
- Year
- 2026
- Stack
- n8n · OpenAI API · Evolution API · PostgreSQL · JavaScript · HTML and CSS
The problem
WhatsApp messages landed on a single phone. Nobody knew which customer was urgent, quotes got lost between chats and the workshop heard about sales through word of mouth.
The solution
An AI agent answers WhatsApp around the clock, asks what it needs and qualifies each customer by size, source and priority. Everything is stored in PostgreSQL and six views turn it into work — Sales quotes and closes, the Workshop completes orders with a PIN, Management controls prices, bundles and discounts, and Metrics tracks conversion.
Highlights
- Conversational agent powered by OpenAI, orchestrated in n8n and connected to WhatsApp through Evolution API.
- Automatic customer qualification with priority alerts for business accounts.
- The full sales flow, from new lead to quote, confirmed purchase and workshop order.
- Discounts that require Management approval, and versioned bundles that prevent conflicting edits.
- Refill and inspection alerts that drive repeat sales.
- Accessible panels — dialogs with managed focus, tables that become cards on phones, and light or dark themes.
Demo entry points
- Metrics Conversion, funnel, customer sources and refill alerts.(opens in a new tab)
- Sales Read the agent's conversation, log a quote and confirm the purchase.(opens in a new tab)
- Workshop Complete a work order with PIN 1111.(opens in a new tab)
- Management Approve a discount, adjust prices and manage bundles.(opens in a new tab)
- Dev Switch the agent on or off and add technicians.(opens in a new tab)
How it's built
The agent. n8n receives every WhatsApp message from Evolution API, loads the customer’s history from PostgreSQL and asks OpenAI for the next reply, with instructions and the current prices and bundles. The agent detects whether the customer is a household or a business, which service they need and when to hand off to a person.
The panels. Six lightweight HTML pages, no framework, backed by n8n webhooks. Each role sees only what it needs, and every action has a single path: log a quote, confirm a purchase or complete an order. Writes are idempotent, so a double click or a network retry never duplicates a sale.
About this demo. It is a copy of the real panels running against an API simulated in the browser. The data is fictional, client information has been removed, and anything you change is kept only in your tab. The panels themselves are in Spanish, as in production.
Screens
