Fin, formerly Intercom, has announced the launch of Fin Voice 2, its next-generation AI phone support agent designed for high-resolution customer service interactions.
Built with Apex Flash, this update delivers a reported 24.5% improvement in resolution rates, responding roughly half a second faster, and is optimized for natural conversations.
This launch highlights Fin’s strategy to build specialized AI for autonomous and scalable customer service outcomes.
Announcing the launch on LinkedIn, Eoghan McCabe, CEO and founder at Fin, argues that voice AI technology has now reached a level where it can reliably deliver the quality needed for mainstream adoption.
“Voice is just extremely hard,” he explained.
“And while we all know that the future of customer experiences will be agent-driven voice, we're not there yet today. That changes today.”
When Conversation Isn’t Enough
As a general-purpose model, the original Fin Voice had been proven effective for conversation but was less optimized for the capabilities required for modern customer support.
Traditionally, early voice AI systems were often judged on their ability to hold natural conversations; however, many businesses are now turning their focus towards operational efficiency.
Whilst a general-purpose model can be highly capable across many domains, customer support now requires a narrower set of skills, and broad models can introduce unwanted variability in a support environment.
Today's customers are less impressed by conversational novelty and now ask whether it can reliably solve customer problems, creating demand for systems optimized for support performance.
Built for Instant Response
The latest version of Fin’s AI-powered phone support agent is built on Apex Flash, a proprietary model developed for customer service voice interactions.
By replacing the general-purpose model, Fin has enabled the agent to move toward specialized support outcomes rather than broad conversational capabilities.
In fact, the change has delivered a 24.5% improvement in resolution rates and reduced response latency by roughly half a second, designed to better understand support requests and generate responses that sound more natural.