Chatbot vs AI agent: which one resolves tickets?
Every support tool now claims to be "AI-powered". The distinction that matters is not the model inside - it is whether the software can finish the ticket. Here is the difference, and the questions that expose it in a demo.
The test: can it do the thing?
A chatbot answers from a script or a knowledge base. Ask "where is my order?" and it links your tracking page. Ask "cancel my order" and it explains how cancellation works - then leaves the customer to do it, or leaves the ticket for your team.
An AI agent closes the loop: it classifies the intent, reads the actual order, takes the action (looks up fulfillment, issues the refund, cancels the unfulfilled order), and resolves the conversation - escalating to a human when confidence is low or the topic is sensitive. The customer asked for an outcome; the agent produces the outcome.
Why the difference shows up in your costs
A chatbot in front of your queue reduces some contacts but leaves the work: your team still opens Shopify, still processes the refund, still writes the closing reply. The ROI of "answers questions" is a fraction of the ROI of "removes tickets".
This is also why billing models diverge. Reply-priced tools happily charge for a deflection that resolved nothing. Resolution-priced agents only earn when the ticket is finished - which forces the vendor to care about the same number you do.
Agents need guardrails chatbots never did
The moment software can issue refunds, safety becomes an engineering requirement, not a policy document. Before connecting a store to any agent, make the vendor show you:
- Grounding with citations - every answer traceable to your policy or live data, so it cannot improvise a return window.
- Action allow-lists and thresholds - refunds and cancellations restricted by intent, amount, and confidence; sensitive topics always routed to a human.
- Idempotency - retrying a refund must return the original result, never charge twice.
- An audit trail - what it said, what it did, and why, on every conversation.
- A human-in-the-loop mode - the agent drafts, your team approves, until the drafts earn autonomy.
Questions that expose the difference in a demo
Ask the vendor to cancel a real (test) order end to end, then ask what happens if the same refund request is submitted twice. Ask what the customer sees when the bot is not confident. Ask to replay your own past tickets before you pay. A chatbot vendor will steer the conversation back to deflection rates; an agent vendor will show you the audit log.
See it on your own tickets
Replay real tickets in a sandbox and watch the agent resolve them - before you pay.