AI customer support: what to automate, and what not to
The goal of AI customer support is not to automate everything. It is to automate the right things and hand off the rest — quickly, without the customer noticing.
The wrong goal
Teams often set the wrong success metric for AI customer support: "what percentage of conversations can the AI handle without human intervention?"
This metric incentivises automating conversations that should not be automated — complex complaints, high-stakes decisions, emotionally charged interactions. The result: a customer trying to resolve a billing dispute, getting stonewalled by an AI that refuses to escalate, and writing an angry public review.
A better frame
The right question is: for which conversation types does AI handling lead to better customer outcomes than human handling?
For most support teams, this is 60–70% of volume — transactional, FAQ-shaped, and status-check queries that have narrow, verifiable answers. The AI handles these faster and more consistently than a human.
The remaining 30–40% — complex, emotional, high-stakes, ambiguous — should go to humans. Not because AI cannot generate a response, but because the cost of a wrong AI response here is much higher than the cost of a human handling it.
The handoff is the product
The transition from AI to human is where most AI support products fall down. Bad handoffs look like: the AI says "let me transfer you to a human" with no context, the human agent starts from scratch, the customer repeats everything.
Good handoffs: the AI hands off with a structured summary (what the customer asked, what was tried, what the customer said about the proposed solutions), and the human agent starts informed.
We designed the handoff as a first-class operation. The AI generates a handoff summary before transferring. The agent inbox shows this summary at the top of the conversation. First-reply time from the agent goes down; customer satisfaction goes up.
What to automate first
If you are starting: automate order status, store hours, return policy, and FAQ lookups. These have clear correct answers, low emotional stakes, and high volume. Measure containment (percentage resolved without human) and CSAT (satisfaction score). Only expand automation after you have baseline data.
The fastest path to AI support ROI is not full automation. It is smart triage — letting AI handle what it is good at, and getting out of the way when it is not.