← Back to portfolio
AI / Automation

SlackBot AI: How I built an AI email-to-Slack router in 23 hours

An intelligent email categorisation and channel-routing system. Built in 23 hours. Adopted by 200+ organisations. Reduced team email time by 60%.

23 hoursBuild time
200+Organisations adopted
60%Email time reduction
Slack · n8n · XanoStack

The problem

Teams using Slack were still drowning in email. The workflow was broken: important emails landed in inboxes, people forwarded them manually to Slack channels, context got lost, things fell through the cracks.

The root issue wasn't email volume - it was routing. Every email had a natural destination (a Slack channel, a person, a workflow), but there was no intelligent layer to make that connection automatically.

"We're spending 2–3 hours a day just deciding where to send things. Half our Slack channels are ghost towns because nobody routes email there." - Typical user complaint before SlackBot AI

The opportunity

With LLMs becoming accessible via API, the routing problem was suddenly solvable without a large ML team. The question wasn't whether AI could categorise emails - it clearly could. The question was: how do you build a system that actually deploys in an organisation within minutes, not months?

That's where the product thinking mattered as much as the technology.

The solution

SlackBot AI is a zero-configuration email-to-Slack routing system. It connects to a team's email inbox, uses an LLM to categorise and triage each incoming email, then routes it to the correct Slack channel with full context - subject, sender, summary, and suggested action.

How it works

  1. Email arrives in the monitored inbox
  2. n8n workflow triggers on new email (polling or webhook depending on provider)
  3. Email body + metadata passed to OpenAI for categorisation and summarisation
  4. LLM returns: category, priority, suggested channel, one-line summary, suggested action
  5. Xano stores the log + handles routing rules per organisation
  6. Formatted Slack message posted to the correct channel with full context
  7. Team reacts or replies directly in Slack - email is a ghost from this point

Stack decisions

n8n Xano OpenAI API Slack API Gmail / Outlook webhooks

Why n8n over Make.com or Zapier?

Self-hosted n8n gave us the ability to process email content without routing sensitive data through third-party servers. For enterprise clients with data compliance requirements, this was non-negotiable. n8n's flexibility also let us build custom conditional logic that would have been impossible in Zapier without 50 separate Zaps.

Why Xano for the backend?

We needed per-organisation routing rules, a message log, and a simple API that n8n could call. Xano gave us a production-grade database + REST API layer without writing a single line of backend code. Each organisation had their own routing table - Xano's relational database structure made this trivial to implement.

Why not build a full web app?

The temptation was to build a dashboard for users to configure routing rules. We killed that scope in the first hour. The MVP question was: can an AI make routing decisions without user configuration? If yes, the dashboard is a nice-to-have for v2. If no, the product doesn't work at all. We validated the AI routing first - it worked - and shipped without the dashboard.

Results

200+
orgs adopted
60%
email time saved
23h
build time
~0
false routes in testing

Adoption was organic - teams shared it with each other. The 200+ organisation count came without any marketing spend. The routing accuracy was high enough that teams stopped double-checking the AI within the first week of use.

Key decisions and tradeoffs

Ship in 23 hours or spend 2 weeks building?

The 23-hour constraint was intentional. I set a hard deadline to force scope discipline. Every feature that wasn't essential to the core "email → categorise → route" loop was cut. The result was a product people could install and use in 10 minutes, not configure for a day.

LLM categorisation vs. rule-based routing

Rule-based routing (if subject contains X, send to channel Y) would have been faster to build. But it would have required every organisation to configure their own rules - a setup friction that kills adoption. LLM categorisation meant zero setup: the AI figures out where an email belongs without you telling it. That's the difference between 200 organisations adopting it vs. 20.

What I'd do differently

What this demonstrates

SlackBot AI was built in 23 hours and served 200+ organisations. That's not a lucky accident - it's what happens when you apply product thinking to no-code tooling. The constraint forces you to find the core of the product, and no-code lets you ship that core without a 6-month dev cycle.

This is what I bring to every project: the ability to identify what actually needs to be built, then build it - fast, clean, and production-ready.

Need something like this built?

I ship AI products in days, not months. No-code stack, production-grade output, full product ownership.

Book a 30-min discovery call →
← Back to full portfolio