The 4 building blocks of AI-first escalation
Understand how guidance, attributes, escalation rules, and workflows combine to route conversations to your team with full context.
No matter how capable your AI agent is, some conversations will always need a human. A customer threatening to cancel, a complex billing dispute, a sensitive complaint: these are moments where your team’s judgment matters.
The question is, when those conversations reach your team, do they arrive with full context and a clear path, or does your team spend the first few minutes just figuring out what’s going on?
Prerequisites
Before you start, make sure you have the following in place:
- Team inboxes configured: Your support team should have inboxes set up in Intercom so that escalated conversations have somewhere to land.
- Fin added to a workflow: You don't need Fin live yet, but you should have created a workflow with Fin as the first step. This is where you'll connect your guidance, attributes, and escalation rules.
Video transcript
So every conversation that Fin handles ends in one of two ways. Fin resolves it completely, no humans involved, or Fin escalates it, either because it can't resolve the issue or because we've decided that that type of conversation should always go to our team. Now, traditionally, escalation was reactive. It usually meant customers had to wait while they were transferred from our frontline team to a specialist who could actually help them. But in AI first support, escalation is proactive. That means we decide in advance which conversations need to be handled by our team, and Finn does all the work to get them there. Let's look at this in a bit more detail. So say a customer messages in and says that they wanna cancel their subscription. Fin picks it up and starts by asking a clarifying question. Now this is driven by guidance that we've set up, which tells Finn to gather the context that we know our team will need. Then as the conversation progresses, Fin identifies that this is a cancellation request and that the customer is frustrated. Now it does this using attributes that we've set up. And because we've also designed escalation rules to go with it, Fin knows to hand this kind of conversation off to our team, so that's what it does. Now we can pretty much choreograph any type of escalation experience that we want using a combination of guidance, attributes, and escalation rules, which we then bring together and control through a workflow. So here you can see how we're deploying Fin. And as you can see, it's following our guidance, detecting our attributes, and then using our escalation rules, it's saying, this actually needs to be escalated. And you can see if the issue is related to a cancellation, it's gonna send them this message, assign them to the retention team, and apply a specific retention SLA. Now all of this is happening in the background. From the customer's perspective, it's seamless. They just go from talking to Fin to talking to someone who already knows what's going on. So once the conversation lands in our inbox, our team has everything they need to jump right in. They're not spending the first two minutes just figuring out what the whole conversation is about. So these are the features that make escalation work, and the groundwork that we put in now is gonna pay off with every single conversation that reaches our team. And in the next few lessons, we're gonna show you how to configure them.
Reactive vs. proactive escalation
In traditional support, escalation was reactive. A customer would explain their issue to a frontline agent, wait to be transferred, and then often repeat themselves to someone new. The experience was frustrating for customers and inefficient for teams.
In AI-first support, escalation is proactive. You decide in advance which types of conversations need human attention, and Fin handles the routing automatically. By the time a conversation reaches your team, it arrives with full context, assigned to the right person, with the right priority.
The four building blocks of AI-first escalation
Proactive escalation in Intercom is built on four features that work together. Each one plays a specific role in shaping the handoff experience.
| Building block | What it does | Example |
|---|---|---|
| Guidance | Tells Fin how to behave during a conversation, including what questions to ask and what context to collect before escalating | “If a customer mentions cancellation, ask them for their reason and how long they’ve been a customer” |
| Fin Attributes | Lets Fin identify and tag characteristics of a conversation as it progresses, like topic, sentiment, or intent | Fin detects that a conversation is a cancellation request and the customer is frustrated |
| Escalation Rules | Define the conditions under which Fin should hand a conversation to your team, based on the attributes Fin has detected | “If the issue is cancellation AND sentiment is frustrated, escalate immediately” |
| Workflows | Bring everything together: deploy Fin with your guidance, connect your attributes and rules, and control what happens when a conversation is escalated (team assignment, SLA, messaging) | Cancellation conversations are routed to the retention team with a specific SLA and a personalized message |
How they connect
Think of it as a chain. Guidance shapes how Fin handles the conversation. Attributes identify what kind of conversation it is. Escalation rules decide whether it needs a human. And workflows tie it all together, controlling how Fin is deployed and what happens at the moment of handoff.
The cancellation example from the video shows this in action: Fin asks a clarifying question (guidance), detects a cancellation request from a frustrated customer (attributes), triggers an escalation (rules), and routes the conversation to the retention team with a specific SLA and message (workflow).
The payoff
The work you put into designing your escalation experience pays off with every single conversation that reaches your team. Instead of spending time triaging and gathering context, your team can focus on what they do best: helping customers.
