How to set up Gorgias AI to detect refund vs exchange intent (2025)

Picture this: a message pops up in your Gorgias helpdesk. "This isn't right, I need to return it." Simple enough, right? But what does the customer actually want? Their money back, or just a different size? Guessing wrong kicks off a frustrating back-and-forth that wastes everyone's time and leaves the customer unhappy.

Gorgias AI can help automate these replies, but it doesn't magically know the difference between a refund and an exchange out of the box. You have to teach it. We're going to walk through exactly how to set up your Gorgias AI Agent to tell these two common requests apart. And stick around, because we'll also look at a smarter way to do this that skips a lot of the manual work.

What you’ll need to get started

Alright, before we jump in, let's get our ducks in a row. The setup isn't too complicated, but you'll need a few things to make it work smoothly.

• First, you'll need a Gorgias account with the AI Agent feature.

• You'll also need admin access to Gorgias so you can actually create and change the AI rules.

• Have your return and exchange policy handy and clearly written out. Your AI is only as smart as the rules you give it.

• Make sure Gorgias is hooked up to your ecommerce store (like Shopify, BigCommerce, or Magento) to pull in order info.

A screenshot showing Gorgias's integration settings, which is essential for the Gorgias AI to detect refund vs exchange intent from message text.

A step-by-step guide to configuring Gorgias AI

Here's the thing about Gorgias AI: it doesn't inherently know a refund from an exchange. You have to spell it out. We'll do this using a feature called Guidance, which is basically a rulebook for your AI Agent. Let's get it set up.

Step 1: Create a new guidance for all return-related inquiries

First up, head over to the AI Agent section in your Gorgias dashboard and hit 'Create New Guidance.' This is going to be your main playbook for any ticket that even whispers the words 'return,' 'refund,' or 'exchange.' It's the master document the AI will check every time a customer wants to send an item back.

This image displays the Guidance settings in Gorgias, where you can configure the AI to detect refund vs exchange intent from message text.

Step 2: Write specific instructions for refund intent

Now, inside your new Guidance, you have to tell the AI exactly what to look for. Don't be shy, be specific. The AI is basically just matching keywords and the context you give it. You'll need to write a clear instruction listing all the words and phrases that scream 'refund.'

For example, you could write something like this:

"If the customer's message contains words like 'refund,' 'money back,' 'not happy,' or 'don't want it anymore,' you should identify the intent as a refund request. Apologize for the issue and provide the customer with a direct link to our return policy page and explain the next steps for getting their money back."

Step 3: Add separate instructions to detect exchange intent

Next up, you'll do the same thing for exchanges, but in a separate instruction within the same Guidance. It's important to keep them distinct so the AI doesn't get its wires crossed.

Your exchange instructions could look something like this:

"If the customer's message mentions a 'different size,' 'another color,' 'wrong item,' 'swap,' or 'exchange,' you should identify the intent as an exchange request. Ask the customer for the specific size or color they would like to exchange for and explain our exchange process."

Step 4: Define the automated actions for each path

Okay, so the AI has figured out what the customer wants. Now what? It needs to actually do something. In Gorgias, you set up 'Actions' for this. For returns and exchanges, this usually means sending a specific reply or flagging the ticket for a human.

For a refund, you might have the AI automatically reply with a link to your returns portal (if you use something like Loop Returns) or just give them the shipping address.

For an exchange, the action could be to ask a follow-up question like, "Sure, what size did you need instead?" before looping in a human agent to finish the job.

Just a heads-up, this method is pretty rigid. If a customer says something you haven't thought of and added to your keyword list, the AI will probably get stuck, and one of your human agents will have to jump in anyway.

A view of the Gorgias action sequence setup, a key step when using Gorgias AI to detect refund vs exchange intent from message text.

Step 5: Test and deploy your new guidance

Before you unleash this on your customers, you'll want to test it. Gorgias has a 'Test Mode' for this. You can type in different customer messages and see exactly how your AI Agent will respond based on the rules you just wrote. It's a great way to see if your keywords are actually working.

Try throwing a few different scenarios at it: a straightforward "I need a refund for order #12345," an obvious "Can I exchange this for a large?," and maybe a tricky one like "This isn't what I expected" to see how it handles ambiguity.

When you're happy with the results, go ahead and activate the Guidance. But don't just set it and forget it. Keep an eye on how it's doing. You'll probably need to pop back in and add new keywords as you see the different ways real customers ask for things.

The challenge: Why rule-based AI falls short

Setting up Guidance in Gorgias is a decent start, but you've probably noticed it's a lot of manual work. You're basically building a giant 'if this, then that' list of keywords. And honestly, that approach has some real drawbacks.

It breaks easily

Imagine a customer writes, "This shirt is way too big, can you send me the next size down?" They clearly want an exchange, but because they didn't use the word 'exchange,' your keyword-based AI might just throw up its hands and default to a generic refund response.

It’s a maintenance headache

You have to babysit it. You're always on the lookout for new phrases customers are using, constantly updating your Guidance. It turns into a never-ending task for your team.

It doesn't actually understand anything

The AI isn't thinking; it's just matching words. It can't handle anything complex. What about a message like, "I'd like to return the blue shirt for a refund, but can I exchange the red one for a medium?" A keyword system will completely short-circuit on that.

For teams that want to automate more with less effort, a smarter approach is needed, one that learns from your team's expertise instead of just following a script.

A better way: Use an AI that learns from your past tickets

So what's the alternative to manually writing rules until your fingers fall off? Using an AI that learns from your team's past work. Platforms like eesel AI are built to do just that by digging into your historical support tickets.

Think about it: an AI that reads through your last 10,000 resolved tickets in Gorgias, learns exactly how your best agents handled thousands of different refund and exchange requests, and then just does that. Automatically. That’s the fundamental difference.

Let the AI detect intent from your data, not just your rules

With a tool like eesel AI, you can ditch the endless keyword lists. The AI trains on your past tickets, so it picks up on all the subtle ways your customers talk. It knows "send me a smaller one" is an exchange request because it's seen your agents treat it that way hundreds of times. This makes it way more accurate and a lot less fragile than a simple keyword matcher.

Test with confidence using real-world simulations

But how do you trust it's ready to go live? Instead of just testing one or two phrases, eesel AI gives you a full-blown simulation mode. You can run the AI against thousands of your actual past tickets in a totally safe sandbox. You'll get a clear report showing how it would have responded, how many refunds vs. exchanges it got right, and even how much time and money you could save. You can launch it knowing exactly what to expect.

Build flexible workflows, not rigid rules

And let's be real, handling a return is more than just sending one reply. Sometimes you need to check inventory, generate a discount code, or tag the ticket for a manager to review. The workflow engine in eesel AI lets you build custom, multi-step automations that can connect to your other tools through an API. This means you can build resolutions that actually solve the whole problem, not just fire off a canned response.

From manual rules to intelligent automation

Getting your Gorgias AI set up to tell refunds from exchanges is a great first step into automation. Following the steps we laid out will give the AI a clear rulebook, its Guidance, to follow, which can definitely help handle some of the simple tickets and take a bit of work off your agents' plates.

But this rule-based method is really just scratching the surface. The real power comes from AI that can learn from the collective knowledge your team has built up over thousands of tickets. When you switch from writing manual rules to using an AI that understands how your customers actually talk, you get more accuracy, less busywork, and automations that can handle the entire problem from start to finish.

If you're curious to see what an AI trained on your own data could do for your team, why not try eesel AI and run a simulation on your Gorgias tickets?