AI Agents for Customer Support
AI Customer Support Agents That Act Within Your Policy
Expedey builds AI support agents that answer from your help center and order data, take actions such as status lookups and refunds only within your policy, and hand off to a person with the full context when they should.
Our open-source agent harness template includes a working order-support agent: policy enforced in the tools, approval before refunds, a hand-off to a person and an evaluation set with a prompt-injection test. You can read the code before you talk to us.
What a Support Agent Can Do
Start with questions and lookups, then add actions one at a time as the agent proves itself on your evaluation set.
Answer from your help center
Searches your articles and policies and answers in plain language, instead of guessing.
Look up orders and accounts
Checks order status, delivery dates and account details through read-only tools.
Take actions within policy
Refunds, cancellations or address changes, checked against your rules in code, not just in the prompt.
Ask before risky actions
Anything that moves money or changes customer data waits for a person to approve it.
Hand off with context
Escalates angry customers, edge cases and anything outside policy to your team with the full story.
Improve from real conversations
Conversations that went wrong become new evaluation cases, so the same mistake is caught next time.
How We Keep Support Agents Safe
A support agent talks to anyone who opens the chat, so the limits have to hold even when a customer tries to talk it out of them.
Policy enforced in the tools
Refund windows and limits are checked in code before anything happens, so the model is never the only check.
Checks before approval
Ineligible actions are rejected before anyone is asked, so people only approve things that can actually happen.
Prompt-injection tests
Customer messages are treated as data. The evaluation set includes attempts to trick the agent into acting.
Evaluations before launch
Real questions with the expected behaviour, scored on every prompt or model change.
A trace for every conversation
Every tool call, approval and answer is recorded, so you can see exactly why the agent did what it did.
Cost per conversation
Token budgets on every run and a running cost per resolved conversation.
How an AI Agent Project Runs
- 01
Scoping call
We pick one workflow with a clear owner, a measurable outcome and data we can access. If an agent is the wrong tool, we say so.
- 02
Pilot on your data
Usually 2–4 weeks. A working agent on real examples, read-only where possible, plus an evaluation set that defines what good looks like.
- 03
Hardening
Integrations, permissions, approval steps, failure handling and monitoring, tested against the evaluation set.
- 04
Production launch
A staged rollout: shadow mode first, then a share of real traffic, then full volume once the numbers hold.
- 05
Run and improve
Ongoing monitoring, re-evaluation when models change, and new tools or workflows as the agent earns trust.
Engagement Models and Pricing
Every engagement starts with a free scoping call. Prices are in US dollars.
AI pilot
From $5,000
One workflow, built on your data in about 2–4 weeks, with an evaluation set and a go/no-go report.
Production build
Quoted after the pilot
Fixed price or time and materials. Integrations, approval steps, monitoring and rollout, scoped from what the pilot proves.
Dedicated AI team
$25–49 per hour
AI engineers who work as part of your team, billed monthly, for ongoing agent and product work.
AI Customer Support Agents: Frequently Asked Questions
What is an AI customer support agent?
An AI customer support agent uses a large language model to answer customer questions from your help center and data, and to take actions such as looking up an order or issuing a refund through tools you control. Unlike a scripted chatbot, it can handle questions phrased in many ways and complete multi-step requests.
Will an AI agent replace our support team?
It handles the routine questions and lookups and hands everything else to your team with the context gathered so far. Your team keeps the complex cases and approves risky actions.
How do you stop an AI agent from giving refunds it shouldn't?
The refund rules are enforced in the refund tool itself, not only in the prompt. Ineligible refunds are rejected before anyone is asked, eligible ones wait for a person's approval, and the tool checks the rules again before it acts.
What about prompt injection, when a customer tries to trick the agent?
Customer messages are treated as data, never as instructions that change the agent's rules, and every action is limited by checks in code. Our evaluation sets include injection attempts, such as a customer claiming to be an admin and asking for a refund.
Have you built a customer support agent we can look at?
Yes. Our open-source agent harness template on GitHub (https://github.com/Expedey/agent-harness-template) includes an order-support agent with order lookups, help-center search, refunds behind approval, a hand-off tool and ten evaluation cases.
How much does an AI customer support agent cost?
A pilot on your own tickets and help center starts at $5,000 and usually takes 2–4 weeks. Production cost depends on the systems the agent connects to and the actions it can take.
Get in touch
Tell us about your project, whether it's an AI agent, an MVP or a web or mobile app. Email us and our team will get back to you.
info@expedey.com+1 (346) 469-4393
Our Offices
USA
12828 Willow Centre Dr Ste D #978 Houston, TX 77066
