How to get Intercom Fin AI to hand off to a human after a confidence threshold

We’ve all been there: stuck in a loop with a support bot that just isn’t getting it. AI agents can be incredibly helpful for customer support, but their real value isn't just in answering questions, it's in knowing when to step aside and let a person take over.

Getting that handoff right is what separates a helpful automated experience from a frustrating one. When your AI understands its own limits, customers feel heard instead of ignored. This guide will walk you through, step by step, how to set up rules so Intercom Fin AI will hand off to a human after its confidence threshold is met. Doing this well keeps customers happy, prevents them from leaving, and frees up your team to handle the problems that actually need their attention.

What you'll need

Before we get started, let’s make sure you have everything in place. You’ll need an active Intercom subscription that includes the Fin AI Agent add-on. You'll also need admin access to your Intercom workspace to get to the settings we'll be discussing. It's also really helpful to have a decent idea of your most common support questions and which ones almost always end up needing a human touch.

How Intercom Fin's handoff rules work

So, how does Intercom know when to give up? You might be picturing a simple "confidence score" slider you can set from 0 to 100, but it’s a little more involved than that. Fin’s decision to pass a conversation to a human isn't based on a single score, but on how well it understands your help articles and the specific rules you give it.

Here’s a peek at what’s happening behind the scenes. When a customer asks a question, Fin scans your connected knowledge base for a direct, clear answer. If the information is fuzzy or it can't find a close match, it recognizes that it might not have the right solution. This is your cue to step in and provide direction.

The main feature Intercom gives you for this is called "Guidance." This is where you tell Fin how to act, what topics it shouldn't touch, and when to call for help. Think of Guidance as the playbook you give your AI, with rules that act like a confidence threshold. You're essentially telling it, "If a question is about X, Y, or Z, don't even try to answer, just pass it to the team."

A step-by-step guide to setting up handoff rules

Alright, let's jump into your Intercom account and get this set up.

Step 1: Identify escalation triggers

Before you click a single button, you need a plan. The best AI handoff rules are based on knowing what your customers actually need. Spend some time digging through your recent support tickets and find the common themes that should always be handled by a person.

Here are a few good places to start:

• Money or personal info. Anything involving billing, refunds, personal data, or security concerns should go straight to a human. No exceptions.

• High-value customers. You might want any question from a VIP or enterprise account to get a personal touch from an agent right away.

• Frustrated customers. Look for keywords that signal someone is getting angry or is stuck on a problem. Phrases like "still not working," "frustrated," "cancel," or "this is unbelievable" are pretty clear signs that a person needs to intervene.

• Complicated technical problems. If an issue needs detailed troubleshooting that isn’t covered in a standard help doc, it’s better to send it directly to your tech support experts.

Step 2: Create escalation rules in Intercom

Now that you know your triggers, it’s time to teach them to Fin.

You'll do this in the Fin AI settings inside your Intercom workspace. Head over to Fin AI Agent > Train > Guidance.

In this section, you can create new rules using plain English. For example, to make sure all billing questions get to your team, you could create a rule like this:

"If a customer mentions 'refund', 'billing issue', or 'payment problem', immediately hand the conversation over to the support team."

You can pack multiple keywords and phrases into one rule. The trick is to be as specific as you can. These rules only work if you can predict how your customers talk. If you tell Fin to look for "billing issue" but your customers usually say "invoice problem," your rule isn't going to work. This means you'll need to check in on these rules from time to time and update them based on real customer conversations.

Step 3: Design the handover experience

So, Fin has decided to pass the conversation on. What does that look like for the customer? A smooth handoff is just as important as making the decision in the first place. You can tweak these settings in your Workflows or under the Fin > Deploy section.

Here, you can decide what happens next. A good approach is to route the chat to a specific team inbox, like "Billing Team" or "Technical Support," so the right person sees it.

It's also a good idea to set a friendly handoff message so the customer knows what's happening. Something like, "I can't quite solve this myself, so I'm passing you over to our specialist team who can take a closer look for you," sets expectations and makes the transition feel natural.

Step 4: Test your handoff logic

You wouldn't launch a new feature without testing it, and the same goes for automation rules. Intercom gives you a way to test your setup before it goes live.

Go to the Fin AI Agent > Test area. Here, you can act like a customer and type in questions that should set off your new escalation rules. For instance, try typing "I want a refund" and check if Fin starts the handover you just configured.

This is helpful for a quick check, but it's a manual process, one question at a time. It won't tell you how the AI will perform across thousands of real-world chats, and you can't run it against your past ticket history. This can leave you guessing about how effective your rules will be, making it tough to roll them out with confidence.

Common mistakes to avoid

Building the rules is one thing; getting them to work well is another. Here are a few common traps that can lead to a messy customer experience if you're not paying attention.

• Making rules that are too broad. A rule like "escalate on negative words" seems smart, but it can trigger way too often. You'll end up flooding your team with chats the AI could have probably solved. Be specific with keywords that strongly point to a need for a human.

• Forgetting a fallback plan. What happens if a chat gets sent to your billing team, but everyone has gone home for the day? Make sure your routing has a plan for after-hours, so customers aren't left waiting in an empty inbox.

• Setting it and forgetting it. Don't just turn your rules on and walk away. Regularly look at the conversations where Fin escalated. Even more important, look for ones where it didn't escalate but should have. This is the only way you'll find the gaps and make your rules better over time.

• Creating a clunky transition. The handoff from AI to human should feel smooth, not like starting a whole new conversation. Make sure your agents can see the entire AI chat history so the customer doesn't have to repeat everything they just typed.

A simpler, more powerful way to control AI handoffs with eesel AI

Setting up these rules in Intercom is a solid start, but managing them can feel like a constant game of whack-a-mole. What if you could test your entire AI setup on past tickets, get a clear forecast of how it will perform, and have more precise control without all the guesswork?

This is where eesel AI can help. It connects directly with your existing helpdesk, including Intercom, giving you more powerful AI features without needing to switch platforms or migrate any data.

Get set up in minutes

Instead of spending hours writing and tweaking text-based rules, eesel AI offers a refreshingly simple, self-serve setup. You can connect your helpdesk, and your AI agent can be ready to go in just a few minutes, learning directly from your past customer conversations.

Test with confidence using simulations

This is what really sets it apart. eesel AI has a simulation mode that runs your proposed AI rules against thousands of your actual historical tickets.

From a single dashboard, you can see exactly which tickets the AI would have answered, which ones it would have passed to your team, and the reason why. You get an accurate, data-backed forecast of your resolution rate before your AI ever talks to a single customer. This takes all the risk out of launching and lets you deploy new automation with certainty.

The eesel AI simulation dashboard showing how AI uses past ticket data to predict future automation rates and handoffs.

Gain granular control with a visual workflow engine

With eesel AI, you're not limited to simple text rules. You get a visual, no-code workflow builder that gives you fine-tuned control over your automation.

You can set up selective automation, deciding exactly which types of tickets, customers, or channels the AI should touch, and then safely escalate everything else. You can also build custom actions. This lets your AI do more than just answer or escalate. It can tag tickets, update fields in your helpdesk, or even pull up order information from an external tool like Shopify before deciding if a handoff is needed.

A screenshot of the visual workflow builder in eesel AI, which provides an alternative to basic rules for the Intercom Fin AI to hand off to human after confidence threshold.

On top of that, eesel AI's predictable pricing is based on features and capacity, not how many resolutions it gets. This means no surprise bills if your handoff rules aren't perfectly tuned, giving you the freedom to experiment and find what works best.

A view of the eesel AI public pricing page, showing transparent, feature-based plans.

Automate smarter, not harder

Making sure your Intercom Fin AI will hand off to a human at the right moment is essential for a good automated support system. It takes a bit of planning, careful rule-making, and a commitment to ongoing testing and tweaking.

While Intercom gives you the basic building blocks, tools like eesel AI offer a more controlled and transparent way to handle the whole process. By letting you run risk-free simulations, build custom workflows, and connect all your knowledge sources, you can stop guessing and start building an AI experience that your customers and your support team will actually appreciate.

Ready to automate with more confidence? Sign up for eesel AI for free and see how you can improve your support automation in minutes.