Automating Customer Support With AI Before You Hire Anyone
How we handle support across a product portfolio with no support team: AI triage, self-serve fixes, and the escalations that still deserve a human.
By FGA Labs
Support is the quiet constraint on running many products with few people. Each product only generates a trickle of tickets, but trickles sum, and context-switching between them is brutal. Hiring a support team for a portfolio of small products never made economic sense. AI finally made a third option real — but only once we stopped treating it as a chatbot and started treating it as a system.
First, delete the ticket
The best support automation is upstream of support. Every recurring question is a product defect wearing a costume: confusing copy, a missing empty-state, an error message that describes the problem without the fix. We review support themes monthly and fix the product until the question stops arriving. AI made this loop faster too — clustering months of tickets into themes is now a one-prompt job instead of an afternoon of spreadsheet archaeology.
What the AI layer actually does
- Answers from real sources — docs, changelogs, known issues — with citations, not vibes.
- Takes safe actions directly: resend a receipt, check an account state, re-trigger an export.
- Triages everything else: severity, product, and a drafted reply waiting for human review.
- Escalates on clear tripwires — billing disputes, data concerns, anger, and anything it is unsure about.
The line we will not automate
Some conversations are the product: a churning customer, a confused new user ready to give up, a bug report from someone who cares enough to write three paragraphs. Automating those saves minutes and costs the only relationship a small product has. The goal of the AI layer is not zero human support. It is spending our scarce human attention exclusively on the conversations where a human changes the outcome.
AI support done well is invisible: customers just notice that answers are fast, accurate, and that a human shows up exactly when one should.
The result across the portfolio is support that scales with products, not headcount — and a founder inbox that contains only the conversations that deserve a founder.