The job
Support teams drown in the same questions over and over. How do I reset my password? Where's my invoice? Why won't this thing connect? Every one of those eats a human's time, and hiring more agents just to answer the same ten questions forever is a terrible use of money. Intercom watched this problem from inside their own help desk software and decided to build the answer themselves. So basically, Fin exists so a company's support team stops paying humans to be human search engines. The AI handles the routine flood, and people handle the stuff that actually needs a brain and a heart.
The agent
Fin started as a support copilot in 2023 and grew into a full AI agent. It learns from a company's help docs, past conversations, and knowledge base, then answers customers in chat, email, and voice — in 45 languages — adjusting to the company's tone. Here's where it gets interesting: Intercom now trains its own small purpose-built model, Fin Apex 1.0, instead of just renting someone else's. They claim it beats the big frontier models on support resolution at a fraction of the cost. You pay $0.99 per resolved conversation, not per seat, so you only pay when it actually works.
Results
The resolution-rate climb is the story here. When Fin launched in June 2023 it resolved about 23-27% of conversations on its own. By late 2025 the average across 6,000+ customers hit 66%, with top customers at 80-90% (Intercom's figures, reported by Tech.eu, Oct 2025). Fin now handles over two million customer conversations a week. VentureBeat reported (March 2026) that the Fin Apex 1.0 model hit a 73.1% resolution rate in Intercom's benchmarks. Named customers include Anthropic, Vanta, Clay, Whoop, Tado, and Miro. Resolution numbers are vendor-reported, so treat them as the company's best-case figures — but the trend line across two years is real.
The catch
Fin is only as good as what you feed it. Companies with messy, outdated, or thin help docs get a confident-sounding agent that answers from bad information — and you find out when customers complain. The per-resolution pricing looks cheap until volume explodes; $0.99 adds up fast at a million conversations. And resolution rate isn't satisfaction rate. A ticket can be "resolved" without the customer being happy about how it was resolved. You also have to do real change management — teams that just flip it on without tuning get middling results, while teams that train and refine it see the big numbers.
Who it's for — who should skip
If you're already on Intercom with a solid knowledge base and a high volume of repetitive questions, Fin is close to a no-brainer — deployment takes weeks, not quarters. Small teams with fewer than a few hundred tickets a month should skip it; the tuning effort won't pay back. And if your product's support is mostly weird, one-off, or emotionally charged problems, no AI agent fixes that. Fix your help docs first. Then let an agent like Fin do the reading out loud.