Most enterprise voice AI runs on frontier models designed for breadth, not the millisecond-level latency and procedural accuracy that phone support actually demands. Fin is taking a different route.
Fin, the customer agent company formerly known as Intercom, has launched Fin Voice 2 — powered by Apex Flash, a model trained specifically for latency-sensitive customer service interactions rather than adapted from a general-purpose foundation.
The performance difference is measurable. Against its predecessor Fin Voice 1, the new model delivers a 24.5% increase in resolution rates and cuts response times by 0.43 seconds. Those figures matter in phone support, where hesitation and mis-resolution are the two fastest routes to customer churn.
The capability shift goes beyond speed. Fin Voice 2 can take action during a call — processing a refund, updating a subscription — rather than routing customers to a human for any transaction that extends beyond information retrieval. End-to-end resolution without a transfer is the design intent, not a stretch goal.
Operational transparency is built into the release too. Real-time Insights surfaces conversations Fin couldn't resolve and generates prioritised fixes that teams can apply with a single click, giving support operations a continuous feedback loop on what the AI is still missing.
Fin launched the product on June 5, 2026 from London. The company positions Apex Flash as the first in a line of task-specific models — a contrast to the industry default of routing phone queries through the same foundation model that handles code generation or essay drafting.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
