Case Study: Intercom cuts Fin AI's time to first token 60% with Honeycomb

Key results

The challenge

Intercom operates Fin, an AI agent for frontline customer support, with hundreds of engineers across 20 engineering teams shipping around 100 changes per day. Understanding and optimizing latency in a high-throughput AI system required deep observability.

The solution

Intercom used Honeycomb to observe and operate Fin.ai, tracing latency contributors such as time to first token and iterating on targeted optimizations.

The results, in context

Intercom reported a 60% reduction in median time to first token, bringing it below eight seconds as of March 2025, with a single optimization shaving two seconds off that median. Teams ship roughly 100 individual changes per day.

Products used