AI Agent vs Chatbot for Customer Support: What Each Can Do

A chatbot answers a customer's question; an AI agent takes care of the request behind it. Ask a chatbot "where is my order?" and it sends a link to the tracking page; ask an AI agent and it confirms who you are, looks up your order and tells you where the parcel is.

Both live in the same chat window, both are built on similar language models today, and many vendors call both "AI". So the difference is easy to miss in a demo and hard to miss once customers start using it. This post sets out what each one can and can't do, walks through five everyday store requests side by side, and ends with the questions to ask before you buy either.

• Chatbot: Software that holds a conversation and answers questions, either from a script you build (button menus and decision trees) or, in newer versions, from your help center and website. It talks. When the customer needs something done, it collects the details and passes them to your team. See chatbot.
• AI agent: AI that plans steps toward a goal and uses tools to take them. In support, the tools are your store systems: orders, returns, subscriptions, customer records. It talks and acts, within rules you set, and hands off when your rules say a person decides. See AI agent.

A useful test: after the conversation ends, has anything changed in your store? If the answer is always no, you have a chatbot, however clever its answers are.

Chatbot: links to your order status page, or asks for an order number and email and passes them to your team. AI agent: confirms who it is talking to (for example, with a code sent to the customer's email), looks up the order, and gives the status and tracking link in the same conversation. This is the most common question most stores get, so it decides much of the difference on its own.

Chatbot: explains the return policy and links to the returns page. AI agent: lists which items in the order can still be returned, asks which ones and why, and opens the return. Whether the agent can also send the money back is a policy choice. Many stores, including those on Concierge, keep that step with a person: the agent opens the return and a teammate approves the refund. See a returns policy your AI agent can follow.

Chatbot: explains how to log in to the customer portal, or opens a ticket. AI agent: finds the subscription, skips the next charge and tells the customer the new date. For subscription stores this one request can be a large share of the inbox, and a ticket that sits overnight can turn into a cancellation.

Chatbot: a good one answers the sizing part from your size guide and links to shipping rates. AI agent: answers from the same guide and your shipping policy, and can check stock of the size asked about. This is the request where a modern chatbot comes closest, because it is mostly a question.

Chatbot: apologizes and offers a contact form. AI agent: recognizes an upset, repeat problem, stops trying to fix it alone and hands the customer to a person with the order history and the earlier complaints attached. Good agents are judged here as much as anywhere: knowing when not to act. See AI escalation and human handoff.

Chatbot vendors often report deflection: the share of conversations that never reached a person. AI agent vendors tend to report resolution: the share where the customer's request was actually taken care of. The numbers can look alike and mean opposite things. A customer who gives up on a chatbot and leaves counts as deflected. A customer whose subscription was skipped in the chat counts as resolved.

When you compare tools, ask how each one counts. Then read 20 conversations it marked as successful and decide for yourself. Our AI resolution rate entry explains the measure, and the AI customer service QA checklist shows what to look for when you read them.

• Most of your questions are about products and policies, not about a specific order.
• You don't sell subscriptions, and returns are rare.
• You get few enough conversations that a person can take every request that needs action within a few hours.
• Your customers mostly write in through website chat, not email, texts or calls.
• You want the lowest monthly cost and can live with a contact form for everything else.

If that describes your store, a chatbot that answers from your help center is a sensible first step. Shopify's free Shopify Inbox may be all you need for a while.

• A large share of your inbox is "where is my order", returns and subscription changes.
• Customers reach you on several channels, including phone and texts, and expect the same answer on each.
• Requests arrive at night and on weekends, when nobody is there to act on them.
• Your team spends its day clicking through the same admin screens for each request.
• You also want help behind the scenes: reports, order checks, follow-ups.

An agent that can act can also act wrongly. That is a reason to set it up carefully, not a reason to avoid it. Four controls matter most:

• Approvals: Decide which actions wait for a person, such as refunds or cancelling a large order. Start strict. See how to keep a human in the loop and human in the loop.
• Verification: The agent must confirm who it is talking to before it shares order details or changes anything.
• Approved sources: It answers from the policies and product facts you approve, and says when it doesn't know. See AI guardrails.
• A record of every step: You can see what it read and what it did in any conversation, so you can correct it.

• Which store actions can it take, and which can I hold back for approval?
• How does it confirm who the customer is?
• Which channels does it answer, and does it remember a customer across them?
• What happens when it can't help: who gets the conversation, and what do they see?
• How does it count success: deflection or resolution? Can I read the conversations?
• How is it priced: per seat, per resolution, or on credits? What does a month like mine cost?

Concierge is on the agent side of this line. It answers customers on website chat, phone, texts, WhatsApp and email, with one memory of each customer across all of them. It tracks orders after confirming the customer, opens returns for a teammate to approve the refund, cancels orders when your rules allow, and pauses, skips or swaps subscriptions. Governed answers keep it to the policies you approve, its own ideas are offered first, and human handoff brings in a person with the whole story. The same agents also do work for your team, like the morning sales report.

Concierge Starter is $20 a month with 20,000 credits; Pro is $199 a month, or $1,990 a year, with 199,000 credits, for the whole workspace. See the AI customer service agent page for what it does alone and what it checks with you first. Glossary: chatbot, AI agent, conversational AI, virtual assistant and AI resolution rate. Read next: what is an AI employee?, conversational AI for ecommerce and how to roll out an AI support agent in four stages.

A chatbot answers questions in a conversation; an AI agent also takes care of the request by using your store's tools, such as looking up an order or skipping a subscription. The chatbot passes the work to your team, while the agent finishes it within your rules.

On its own, ChatGPT is a chat assistant for the person using it. It acts more like an agent when it is connected to tools, such as a store's data through a connector. Neither makes it a customer-facing support agent on your site.

AI agents are safe with customers when you set approvals, customer verification and approved answer sources before going live, and read their conversations in the first weeks. Keep money and policy exceptions with a person.

For stores whose inbox is mostly order, return and subscription requests, an agent does everything a chatbot does and more. For stores with mostly product questions and low volume, a chatbot can still be enough.

An AI agent usually costs more per month than a chatbot, though pricing models vary: per seat, per resolution or on credits. Compare the cost of the requests each one actually closes, including the work a chatbot passes back to your team. See what an AI employee costs.