AI agents in customer service: what to automate and what to keep human

AI agents have changed the economics of ecommerce customer service. Routine questions that once took a human several minutes can now be resolved in seconds, at any hour, for a fraction of the cost. But brands that automate everything tend to learn the hard way that some conversations need a person.

Start with your contact reasons

Before choosing tools, tag a few weeks of tickets by reason. In most apparel and lifestyle brands, a handful of reasons dominate: where is my order, returns and exchanges, sizing and fit, address changes and cancellations, discount code problems. These are predictable, policy-driven and data-backed, which makes them good candidates for automation.

What AI handles well

• Order status, using live order and tracking data
• Starting returns and exchanges within policy
• Sizing questions, using size charts and fit notes
• Address changes and cancellations before fulfillment
• Policy questions: shipping times, return windows, care

What should stay human

• Damaged, defective or unsafe products
• Refunds or credits outside policy
• Chargebacks and suspected fraud
• Angry, upset or vulnerable customers
• VIPs, press, influencers and wholesale accounts

Design the handoff

The handoff is where most hybrid models fail. When the AI escalates, the human agent should see the full conversation, order details and the reason for escalation, so the customer never repeats themselves. Customers should always be able to ask for a person.

Measure quality, not just deflection

Deflection rate on its own rewards an AI for ending conversations, not solving them. Track resolution confirmed by the customer, CSAT for AI and human conversations separately, re-contact within seven days, and a weekly QA sample scored for accuracy and tone.

Use support as an insight engine

Every contact is a signal. Tagging by reason and product shows which styles generate fit complaints, which shipping messages confuse people, and which policies create friction. Fixing those at the source reduces contacts more than any automation.

Our Customer Experience and AI Agents service runs this hybrid model across email, chat, social, voice and TikTok Live.