Intercom went from skeptics to believers on AI

Brian Scanlan, Senior Principal Engineer at Intercom, admits something unusual for a tech company: they were AI skeptics. Intercom had been doing machine learning for years—summarization, auto-responses—but when GPT hit, their initial reaction was "this seems like a fad."

Then they actually tried it on customer support data. The results were extraordinary. Today, Intercom's AI agent, Fin, handles 86% of customer conversations in some deployments. Not assists. Handles.

The overnight change in quality

When Intercom first plugged their customer support data into GPT, the quality of AI-generated responses jumped dramatically. Not incrementally—dramatically. The team went from skeptics to believers in about a week.

The key insight was that customer support is a near-perfect use case for LLMs. Every company has a knowledge base of help articles. Customers ask questions that usually have answers somewhere in that knowledge base. LLMs are excellent at retrieval and synthesis. The matching was obvious in retrospect.

What 86% resolution actually means

When Brian says Fin resolves 86% of conversations, he means end-to-end resolution without human involvement. The customer asks a question, Fin provides the answer, the customer is satisfied, the conversation ends. No escalation, no handoff, no human in the loop.

This isn't cherry-picked data. This is the actual resolution rate at their best customer deployments. The typical deployment sees lower numbers, but 50-70% AI resolution is common. That's still transformative for support teams.

The human role is changing

Intercom isn't building AI to eliminate support teams. They're building AI to change what support teams do. Instead of answering the same questions hundreds of times, human agents handle the complex, nuanced, relationship-building conversations that AI can't do well.

Brian's framework: AI handles the repetitive work at scale, humans handle the work that requires judgment and empathy. This is better for customers (faster responses) and better for support agents (more interesting work).

The investment bet

Intercom went all-in on AI after becoming believers. A huge percentage of their engineering team now works on AI features. They've reorganized product teams around AI capabilities. It's not a feature—it's the strategy.

The bet is that customer support will be fundamentally transformed by AI, and Intercom wants to be the platform that powers that transformation. Early results suggest the bet is paying off: they've signed over 40,000 Fin customers since launch.

This interview was conducted at AWS re:Invent 2025.