How AI Agents Can Reduce Customer Support Work (2026)

A support queue is not “chat.” It is people asking for an order fact, a policy fact, or an exception the policy does not cover. If a bot answers from a help article while the order system says cancelled, you did not reduce the work. You automated a wrong answer in a friendly tone.

Gorgias, reported via Redo, puts where-is-my-order — often shortened to WISMO — at about 18% of incoming requests, and the most common type. That is the usual first slice. This note is the short version: what an agent can take off associates this week, and what still waits.

The job. Stop the associate retyping tracking, the policy version, and “where is my order” for the tenth time today.

This week. Look up the order, the shipment, the product, and the published return policy. Cite the record. Hand off with the evidence attached.

A person still signs. Refunds, cancellations, address changes, gift cards, chargebacks, and “it says delivered and I don’t have it.”

Skip it when a template already closes the ticket, when there is no human queue, or when the only goal is an automatic refund.

This is part 2 of AI Agents for Business. The long build is AI Customer Support Agent for eCommerce. The commercial page is Customer Support Agent.

Our take: week one is lookups plus a handoff. Associates still handle refunds. That feels slower than a bot that “just refunds.” The alternative is an automated refund machine with a chat window.

Copy the allow-list — The lookups, the forbidden writes, and a permission sketch live in support-agent-tool-policy.md. Ship gates: monday-checklist.md. Series index: ecommerce-ai-agents-series/.

FactualMinds is an AWS Select Tier Services Partner. We do not have a published agent case study that claims tickets fell by a percentage. The checklist above is the proof of work.

What “less work” means

The associate does not retype the same status. The agent:

• Reads the order, the shipment, the product, and the published return policy.
• Cites what it found — status, last scan, which policy — not something it remembers from training.
• Hands off, with a session id, when the case is an exception.

It does not close chargebacks, invent a delivery window, or issue a refund without a hard block and a person above your dollar cap. Where that person sits is its own note.

This week, and later

What to use instead

• Tickets a status email already closes. Keep the template. Do not add a bot that restates it badly.
• Leadership only wants auto-refund. Refuse the week-one scope. Put a person and a block on writes first, or do not start.
• A help center that cannot look up an order. That is search, not an agent.
• You need the build note. Support agent and where-is-my-order.

If you only do one thing

Write the week-one list on one page: the lookups, the writes that are forbidden, and the cases that always go to a person. If you cannot name those, you are not ready to reduce support work with an agent. You are ready to buy a widget.

For your technical lead

On June 17, 2026, AgentCore Harness reached general availability (What’s New). Agents Classic is in maintenance for new customers after July 30, 2026. A new storefront bot should not start on Classic action groups.

First-party signals we reuse (not store results) — Server-side tools cut median tool round-trip ~180 ms → ~95 ms on a B2B CRM assistant (12 tools, ~8k turns/day) — Gateway post. Support-style AgentCore at 50K sessions/mo ~$791/mo platform plus model (decision guide). Support-shaped pilots have shown roughly 3× the platform spend when a browser tool stayed on for turns that only needed order and shipment lookups — same support post.

Week-one tool names: getOrder, getShipment, policy, help articles. Later, and only with a block: createRefund, cancelOrder, updateAddress, gift cards. Screenshots of the admin screen are not tools.

What broke — A browser tool left on for every support turn. Detection: platform compute dominated the bill, and the lookup record went muddy. Recovery: browser off by default. Add a carrier API. The roughly 3× spend is the published counter-case — support field-guide post.

What to do this week

1. Pull last month’s ticket mix. If where-is-my-order and policy are not a visible slice, pick a different family on the hub.
2. Confirm the order and shipment can be looked up by a named API. A screenshot of the admin screen is not a tool.
3. Name the person who gets the exceptions. No queue means no agent.
4. Open Customer Support Agent. Book a conversation only after the allow-list exists.

What this post doesn’t cover

It does not retell the permission walkthrough or the where-is-my-order state machine. Those stay in the field guide. It does not treat returns-fraud scoring or refund investigation as week-one work. It does not invent a ticket-volume result. We have not published a new support-desk census for this note. 18%, ~180→95 ms, ~$791/mo, and ~3× browser-on spend are the published figures we reuse.

Primary next step: Customer Support Agent.