A practical guide: How to use Intercom Fin AI to trigger refunds with Custom Actions

Let's be honest, processing refunds is one of those tasks that's super important but also a total drag. It’s the kind of repetitive work that can make even the most patient support agent’s eyes glaze over. You have to be meticulous, follow a strict process, and every minute spent on a refund is a minute not spent helping a customer with a more complex, interesting problem.

But what if you could take that entire task off your team's plate? Safely, reliably, and instantly?

That's the promise of AI agents. They can step in and handle these routine jobs, giving your customers immediate answers and freeing up your human agents to focus on what they do best. Intercom has a powerful AI agent called Fin that can do just this.

In this guide, we're going to get practical. I'll walk you through exactly how to set up Intercom Fin AI to trigger refunds with Custom Actions. We'll cover the technical nuts and bolts, but we'll also talk about the hurdles you might face. And, stick around, because I’ll also show you a much simpler and more affordable way to get this done.

An illustration of the Intercom Fin AI agent.

What you'll need to get started

Before you can roll up your sleeves and start building, you need to have a few things lined up. Setting this up in Intercom isn't a simple point-and-click affair; it requires some technical groundwork. Getting these pieces ready beforehand will make the whole process a lot smoother.

• An Intercom account with the Fin AI Agent add-on: This one's a given. Fin is the AI brain we'll be teaching, so you need to have access to it in your Intercom plan.

• An internal API endpoint: This sounds more intimidating than it is. Think of it as a secure, private hotline that Intercom can use to tell your internal system, "Hey, process a refund for this order." This is the bridge between your helpdesk and your payment or order management system (like Stripe, Shopify, or a custom database).

• Access to a developer: Unless you're comfortable writing code and managing APIs yourself, you're going to need an engineer to build and maintain that API endpoint. This is a crucial step for security and making sure the whole thing actually works.

• Your company's refund policies, written down: You can't teach an AI to follow rules that aren't clear. Before you automate anything, you need to have your refund logic crystal clear. For example, can a customer get a refund after 30 days? Is it a full or partial refund? Do you need a reason? The AI will need to check against these rules, so you need to define them first.

Step-by-step: How to configure refund actions with Intercom Fin AI

Alright, let's get into the weeds. Intercom uses a system of Data Connectors and Fin Tasks to let its AI talk to external systems and perform actions. Here’s how you’d piece it all together.

graph TD  

    A[Customer Asks for Refund] --> B{Fin AI: Ask for Order ID};  

    B --> C[Fin AI: Use 'Process Refund' Data Connector];  

    C --> D{API Endpoint: Process Refund};  

    D --> E{Response};  

    E --> F[Success: Inform Customer Refund is Processed];  

    E --> G[Failure: Inform Customer & Escalate to Human Agent];  

Step 1: Get your refund API endpoint ready

First thing's first: you need that secure hotline I mentioned. Your development team will need to create a private API endpoint that can receive a request from Intercom and kick off the refund process in your backend.

This isn't just a simple link. It needs to be built with security in mind. It should be able to:

• Authenticate requests from Intercom securely. You don't want just anyone to be able to trigger refunds. This usually involves an API key or another authentication method to verify that the request is legitimate.

• Accept the necessary data. To process a refund, your system probably needs key info like an "order_id" or "customer_email". The API needs to be set up to receive this information from Intercom.

• Send back a clear response. After trying to process the refund, the API needs to report back. A simple "success" or "failure" message is essential so Fin knows what to tell the customer.

Step 2: Set up a Data Connector in Intercom

Once your API is ready, you need to introduce it to Intercom. This is where Data Connectors come in. A Data Connector is just Intercom's name for a feature that lets Fin communicate with external tools, like your brand-new refund API.

Inside your Intercom settings, you'll configure a new Data Connector. This involves telling Intercom the address of your API endpoint, how to authenticate (by providing the API key, for instance), and what kind of data to send and expect back. Essentially, you're giving Fin a phonebook entry that says, "When you need to process a refund, call this number, use this password, and give them this information."

Step 3: Build a Fin Task to handle the logic

Now for the fun part: teaching the AI what to do. A Fin Task is where you lay out a series of instructions for Fin to follow. The cool, but also tricky, part is that you do this using natural language.

You'll write out a script for Fin, almost like you're training a new support agent. It might look something like this:

1. "When a user asks for a refund, your first step is to ask for their order ID."

2. "Once you have the order ID, use the 'Process Refund' Data Connector we just set up."

3. "If the connector sends back a 'success' message, you should tell the user their refund has been processed and will appear in their account in 5-7 business days."

4. "If the connector sends back a 'failure' message, you need to tell the user you couldn't process the refund right now and smoothly hand the conversation over to a human agent."

Getting these prompts just right takes some work. You have to think about all the different ways a customer might phrase their request and account for edge cases. What if they provide the wrong order number? What if the API is temporarily down? It often takes a fair bit of trial and error to make Fin's behavior reliable.

Step 4: Test, test, and test again

You wouldn't let a new agent handle payments on their first day without training and supervision, right? The same goes for your AI. Before you unleash this on your customers, you need to test it thoroughly.

Intercom has a testing environment where you can simulate conversations and see how Fin reacts to different inputs. Run through as many scenarios as you can think of: successful refunds, failed refunds, confused customers, typos, everything. Once you feel confident that Fin is handling things correctly and not going off the rails, you can deploy the Task and let it start working for you.

The reality check: Challenges with Intercom Fin AI

So, that’s how you do it. It’s definitely powerful, but setting up custom actions in Intercom Fin isn’t exactly a walk in the park. There are a few significant challenges you should be aware of before you dive in.

• The Cost: Intercom's pricing for Fin is based on a per-resolution model: $0.99 for every ticket Fin successfully closes. Every single refund your AI processes will cost you nearly a dollar. If you're a small shop with a handful of refunds a month, maybe that's fine. But if you process hundreds or thousands, those costs can skyrocket and become a real headache for your budget. You might find that the automation ends up being more expensive than just having a person do it.

• The Complexity: As you can see from the steps above, this is a technical project. It requires coordination between your support team and your developers. Someone has to build and maintain the API, and someone has to carefully craft and test the natural language prompts. It's not something a non-technical support manager can easily spin up on their own, which can create bottlenecks and slow things down.

• The Confidence Gap: It's one thing to have an AI answer a simple FAQ. It's another thing entirely to trust it with your customers' money. Because Fin's logic is based on natural language prompts, it can sometimes be hard to predict exactly how it will behave in every single situation. Launching a critical financial task like refunds requires a huge amount of confidence, and getting there can be tough without extensive, real-world testing.

A simpler, more powerful way with eesel AI

What if you could get the same powerful automation but without the high costs, technical hurdles, and uncertainty? This is exactly why we built eesel AI. It plugs right into your existing helpdesk, including Intercom, and gives you a much more straightforward and cost-effective way to automate complex actions.

Get up and running in minutes

Forget about waiting for developers. eesel AI is built to be self-serve. You can connect your Intercom account with a single click and start configuring your AI agent from a simple, intuitive dashboard. No need to book a mandatory sales call or sit through a lengthy demo just to get started.

Build custom actions with total control

With eesel AI's workflow engine, you're in the driver's seat. You can easily set up custom API actions to handle refunds, look up order details, or update user profiles. Instead of wrestling with natural language prompts, you use a clear, visual builder. You also get full control over the AI's personality and can define exactly which tickets it should handle, so it escalates everything else to your team with no guesswork.

A look at eesel AI's visual workflow builder, which simplifies creating custom actions.

Test with real data and zero risk

This is where eesel AI really shines. Before you activate your AI agent, you can run it in a simulation mode. It will process thousands of your past tickets and generate a report showing you exactly how it would have performed. You can see its projected resolution rate and review every single conversation to see what it would have said. This lets you fine-tune the AI's behavior and go live with complete confidence.

The simulation feature in eesel AI, where you can test AI performance with past ticket data before going live.

Ditch the per-resolution fees

We believe you shouldn't be penalized for being successful. eesel AI has simple, predictable pricing plans with a flat monthly fee for a generous number of AI interactions. We never, ever charge per resolution. This means your costs are predictable, and you can automate as much as you want without worrying about a surprise bill at the end of the month.

eesel AI's transparent, flat-rate pricing page, which offers a predictable alternative to per-resolution models.

Tips for successful automation

No matter which tool you end up using, here are a few best practices to keep in mind for a smooth rollout.

• Start small. Don't try to automate your entire support operation on day one. Pick one high-volume, low-complexity task, like a simple refund request. Nail that process first, learn from it, and then expand to more complex workflows.

• Monitor and iterate. Your AI is not a "set it and forget it" tool. Keep an eye on its performance. Good analytics will show you where it's excelling and where it's getting stuck. This feedback is gold, it can point out gaps in your knowledge base or areas where the AI's logic needs a little tweak.

• Always test before you launch. I can't stress this enough. Never roll out a new automation that interacts with customers without testing it in a safe environment first. A simulation mode or sandbox is your best friend for catching weird behavior before it impacts a real person.

• Choose a predictable pricing model. Be careful with per-resolution fees. They sound small, but they can add up in ways that are hard to predict. A flat pricing structure makes it much easier to calculate your return on investment and build a sustainable automation strategy.

A smarter way to automate refunds

Automating tasks like refunds can radically improve your support team's efficiency and give your customers the speedy experience they expect. While tools like Intercom Fin offer a way to achieve this, it's important to weigh the trade-offs in cost, complexity, and control.

The right AI platform should feel like a true partner. It should empower you to build, test, and deploy powerful automations with confidence, without needing a team of developers or an unpredictable budget. By focusing on simplicity, control, and predictable costs, you can unlock the real power of AI for your support team.

Take control of your support automation with eesel AI

Curious to see how easy it can be? Sign up for a free trial of eesel AI and you can launch your first AI agent in just a few minutes.