AI Customer Support Automation

Stop answering the same ticket twice. In this course you build a live support system where an AI agent resolves most tickets, hands the hard ones to a human, and gets smarter every week.

Introduction

You won't watch Mara talk — you'll build. By the final module you will run a support desk where an AI agent answers ~80% of tickets from your own knowledge base, escalates cleanly to a human when it should, and reports deflection and satisfaction numbers you can defend. Everything is built with tools you already have: an LLM API, your docs, and a mailbox or chat widget.

Module 1 — Your Support Reality, Measured

Before automating anything, you measure what support actually costs you today and decide what "good" means.

Learning Objectives

• Export and categorize your last 100 tickets into a working taxonomy
• Measure baseline volume, first-response time, resolution time and repeat-rate
• Define your target deflection rate and the tickets AI must never touch
• Write the one-page support charter your agent will operate under

Lesson 1.1 — The ticket audit

Pull your last 100 support conversations — email, chat, WhatsApp, wherever they live. Paste batches into your AI assistant and have it propose categories: how-to questions, billing, bugs, shipping, refunds, feature requests. You'll find the classic shape: a handful of categories cover most of the volume.

Build a simple sheet: category, count, average time to resolve, and whether the answer already exists somewhere. That last column is the automation goldmine — questions with written answers are the first ones your agent takes over.

Lesson 1.2 — Baseline metrics that matter

Three numbers rule support: first-response time, resolution time, and repeat-contact rate. Compute them from your export — your AI assistant can do the math from timestamps in one prompt.

Write them down. Every improvement you claim later gets measured against this baseline, and investors and customers both love a before/after with real numbers.

Lesson 1.3 — The support charter

Decide the rules of engagement before the robot shows up: which categories the agent may answer alone, which it drafts for human review, and which — legal threats, angry VIPs, payment disputes — go straight to you. Write it as a one-pager.

This charter becomes your agent's system prompt in Module 3, so write it in plain imperative language: "Answer how-to questions from the knowledge base. Never discuss refunds above €50. If the customer is angry, apologize once and hand off."