Zon en Woning, AI / Automation
An AI email assistant that learns with the team
A customer portal that reads incoming email, drafts replies from its own knowledge base and suggests improvements every week. A person approves everything.
- AI
- Next.js
- Supabase
- n8n
3
AI models, selectable per inbox
- Client
- Zon en Woning
- Role
- Product design, AI, development
- Period
- 2026, ongoing
- Platform
- Next.js + Supabase + n8n
The context
Zon en Woning, based in Ede, supplies and installs sun protection. Much of their customer contact runs through email: questions about quotes, delivery, installation and maintenance.
The problem
Many emails look alike, yet each one still takes manual work. The knowledge to answer them well sits with a few people.
Core problem
An AI that sends email on its own is a risk. An AI that doesn't learn from the team's corrections keeps making the same mistakes.
Our approach
We designed the assistant around the person who sends the email, not around the model.
- 1
Incoming email arrives from Outlook through n8n. Overnight, the AI prepares a draft reply for every conversation, based on the knowledge base, templates and instructions.
- 2
The team edits drafts or discusses them with the assistant, and approves them before they are sent.
- 3
Every week the AI compares its drafts with what was actually sent, and suggests changes to the knowledge base, templates or prompt. A person approves those too.
What we built
A customer portal in Next.js with Supabase for data, login and storage. Each inbox can choose which model writes: Claude, GPT or Gemini. A custom MCP connector lets ChatGPT search the email and prepare drafts. The portal supports multiple brands and languages.
Results
The team no longer starts from an empty email but from a draft, and corrections flow back into the knowledge base. How Zon en Woning talks to customers is now written down instead of living in a few heads.
Facing a similar problem?
Tell us what you're running into and we'll show you how we would approach it.