← Back to Projects

Appointment Guardian

AI Agent for Appointment Management · Built Jan–Feb 2026

An AI agent that predicts cancellation risk, sends smart reminders, and manages waitlists. Built by 3 PMs across 3 countries in 4 weeks with zero lines of code.

The Problem

Small businesses like hair salons, beauty studios, dental offices, and physios lose real revenue to no-shows. Most appointment tools do send reminders, but none actually predict the risk of a cancellation before it happens.

My Role

One of 3 PMs leading end-to-end: discovery, PRD, AI evaluation framework, and prototype validation across France, Germany, and Austria.

What We Built

  • Risk Assessment Tool (RAT) powered by Gemini Flash 2.5, with ~90–95% cancellation prediction accuracy
  • 3 n8n workflows handling booking, risk monitoring, and responses
  • Smart reminders at 72h / 48h / 24h via email + Telegram
  • Automatic waitlist management to fill cancelled slots
n8n workflow: new appointment booking handler
Workflow 1, Booking Handler: ingests new appointments via webhook, normalises client data, and writes them to the source of truth.
n8n workflow: Risk Assessment Tool and messaging agent
Workflow 2, Risk Assessment Agent (RAT): scores every upcoming appointment for cancellation risk, then routes high-risk slots to the messaging agent.
n8n workflow: customer message handler across Telegram and email
Workflow 3, Message Handler: parses replies from email and Telegram, classifies them as Confirmed, Cancelled, or Unclear, and triggers the right follow-up.

North Star metric: Revenue Protected (€), not messages sent

Results

  • PRD rated "Strong", with the autonomy matrix called "exemplary"
  • "Relationship-First Logic" positioning flagged as a trust moat competitors can't easily replicate
  • Rated Strong on: Value Translation · PRD-to-Prototype Fidelity · Learning Velocity

Key Learnings

Building an agentic system forces you to think fundamentally differently than in traditional software engineering. In software, you design for predictable outputs. With AI agents, you design for uncertainty because the system reasons, improvises, and sometimes gets it wrong.