A complete guide to Intercom's Fin AI Agent over chat

Everyone is chasing that dream of instant, 24/7 customer support. Let's be honest, it's not just a nice perk anymore; customers expect it. AI chat agents like Intercom's Fin have shown up promising to be the perfect fix: always available, always helpful, and ready to solve problems on the spot.

But as with most things in tech, the promise and the reality can be two different stories. Getting an AI agent up and running is rarely as simple as flipping a switch. This guide is a straightforward, no-fluff look at what it really takes to use the Fin AI agent over chat. We’ll walk through the setup process, its features, and its complicated pricing. We'll also talk about the common roadblocks, like tricky implementation, costs that are all over the place, and the feeling of being locked into one company's way of doing things.

What is the Fin AI agent over chat?

Fin is Intercom’s homegrown AI chatbot, built to handle customer service chats right inside their platform. It’s powered by some of the latest AI models (like GPT-4) and uses a method that lets it search your company’s knowledge base for answers instead of just making them up.

Its main goal is to figure out what customers are asking in normal, everyday language, give them a quick and accurate answer, and know when to hand the conversation over to a person if things get tricky.

A screenshot showing the Intercom AI chatbot interface, which is central to the Fin AI agent over chat experience.

It's important to know that Fin isn't a tool you can just grab and plug into any system you're using. It’s an add-on that's tied directly to the Intercom platform. That tight connection has some benefits, for sure, but it also has a big impact on setup, cost, and how much freedom you have to do things your own way.

How to set up and train the Fin AI agent over chat

Intercom suggests a "Train, Test, and Deploy" model for getting Fin ready. It sounds simple on paper, but each of those stages has a bunch of steps that can eat up a surprising amount of time. Let's break down what's really involved.

Step 1: Teaching Fin what it needs to know

First, you have to feed Fin information. It starts by connecting to your Intercom Help Center articles, which it treats as its main source of truth. You can also give it public website links, internal documents, and other content to build up its knowledge.

This image displays the process of connecting various knowledge sources to train the Fin AI agent over chat.

But here’s the snag. This process is smoothest when all your important info is already living inside Intercom. For most teams, that’s just not realistic. Your company’s real knowledge is probably spread out across Google Docs, Confluence, or other internal wikis. Getting Fin to learn from those outside sources can be a real headache.

You also have to spend time in the "Guidance" feature, where you manually write instructions to teach Fin your brand's tone of voice and how to handle specific situations. This isn't a one-and-done thing; it takes a lot of tweaking to get it right.

Step 2: Checking Fin's homework

Once Fin has some knowledge to work with, Intercom gives you a testing area where you can ask it questions and see what it says. You can review its answers, mark them as good or bad, and give feedback to help it improve.

A screenshot of the testing environment for the Fin AI agent over chat, where users can review and refine its answers.

This is useful for quick checks, but it's a very manual process. You’re testing one question at a time, which doesn't really show you how Fin would handle the thousands of real customer conversations you get. It makes it tough to get a good sense of its potential resolution rate or how much money it might save you before you're fully committed.

Step 3: Letting Fin talk to customers

When you feel ready to go live, you have two main choices: a "Simple deploy" or an "Advanced setup through Workflows."

The simple option lets you set some basic rules, like which customers see the bot and when it should pass a chat to your team. It's fast, but it doesn't give you much control.

For anything more complex, you have to use Intercom's Workflows. This means building out branching logic flows to manage different topics, customer types, or support channels. It can be powerful, but it can also quickly start to feel like you're coding instead of just setting up a support tool.

An example of Intercom's workflow builder, which is necessary for advanced setup of the Fin AI agent over chat.

Key features and limitations of Fin

No tool is perfect, and Fin is no different. It has some definite strengths, especially if your team lives and breathes Intercom, but it also comes with some major limitations you should know about.

What Fin does well

• Works across different channels: You can use Fin for web chat, email, SMS, and even social media like Facebook and Instagram, which keeps the customer experience consistent.

• Plays nicely with Intercom: Since it's a native feature, Fin fits right in with the Intercom inbox, user data, and reporting. Everything feels connected because it's all part of the same system.

• Combines answers: Fin can pull information from a few different articles to put together one complete answer, which is great for more complicated questions.

Where Fin falls short

Despite its good points, Fin's design creates a few big problems for teams that need a bit more flexibility and control.

• It's a walled garden: Fin is built to work best with content that's already inside Intercom. If your team's knowledge is scattered across tools like SharePoint, Notion, or a shared folder of Google Docs, you're in for a lot of copy-pasting to get that information into the bot. This is a huge issue for businesses that don't run their entire operation on one platform. On the other hand, tools like eesel AI are made to unify all your scattered knowledge instantly. It connects to dozens of sources and can even learn from past tickets in other help desks like Zendesk or Freshdesk.

• You can't really test it at scale: As we mentioned, you can ask Fin individual questions, but there's no good way to see how it would have handled thousands of your past support tickets. This means you're basically guessing what your resolution rate and ROI will be before you start paying. In contrast, eesel AI has a simulation mode that does exactly that. You can run the AI over your historical data in a safe environment, see exactly how it would have performed, and get reliable numbers on performance and cost savings. It lets you build confidence and tweak your setup before a single customer ever talks to it.

• Automation rules are rigid: If you want to customize how Fin behaves beyond the basics, you're pushed into that complicated workflow builder. You don't get simple controls to decide which specific ticket types the AI should handle and which it should immediately send to a human. This makes it hard to roll out automation slowly and safely.

Breaking down Fin's pricing

Alright, let's talk about the price tag. This is where things get confusing, and it's one of the

biggest frustrations for many Intercom users

Reddit

. Fin's pricing isn't a single number; it's a two-part system that makes it more expensive and unpredictable than it looks.

First, you have to be on a paid Intercom support plan. These plans start at $39 per agent, per month and go up to $139 per agent, per month, or even higher.

Second, on top of that monthly fee, Intercom charges you $0.99 for every single resolution the AI provides.

An AI "resolution" happens anytime a customer says their issue is solved or just ends the chat without asking for a human. The problem here is obvious: this model leads to unpredictable costs and actually punishes you for success. The better your AI does and the more questions it answers, the more you pay. This makes budgeting a nightmare and can easily wipe out the cost savings you were hoping for.

This per-resolution model is a common pain point for support teams. If you need to know what you're spending each month, an alternative like eesel AI is a much safer option. With its clear, predictable pricing, you get a full suite of AI tools for one flat fee. You'll never get a surprise bill, no matter how many thousands of tickets your AI agent closes.

Is the Fin AI agent over chat the right choice for your chat support?

So, what’s the final verdict on Intercom's Fin? It really depends. If your team is all-in on the Intercom platform, all your knowledge lives there, and you've got the budget for a fluctuating, usage-based bill, Fin can be a powerful and well-integrated tool.

However, for most teams, those benefits come with some heavy baggage. You're looking at a potentially long and complicated setup, a "walled garden" that doesn't play well with outside knowledge, and a pricing model that can easily get out of hand.

For businesses that value flexibility, simplicity, and predictable costs, a dedicated, third-party AI solution is almost always a better long-term bet.

A smarter way to automate your support

eesel AI gives you all the power of a top-tier AI agent without locking you into a closed system. It’s built to be simple, flexible, and effective.

• Go live in minutes, not months: With a straightforward self-serve setup and one-click integrations, you can have a working AI agent ready to go in less time than it takes to drink a cup of coffee.

• Bring all your knowledge together: Connect to your help desk, Google Docs, Confluence, Slack, and dozens of other tools with a single click. No more data silos.

• Test with real confidence: Use the simulation mode to see how the AI would perform on your actual historical data, so you know exactly what to expect.

• Pay one predictable price: Forget about per-resolution fees. Our transparent plans give you unlimited resolutions, so you can grow without worrying about a surprise bill.

Ready to see how easy AI support can be? Start your free eesel AI trial today and build your first AI agent in under five minutes.