Intercom renamed itself Fin after its AI agent, then Salesforce paid $3.6 billion. When a company renames itself after a feature, that feature is the company.
On May 12, Intercom renamed itself Fin. The fifteen-year-old customer messaging company took the name of its AI agent, the product that now accounts for roughly a quarter of its $400 million in annual recurring revenue and virtually all of its growth. Five weeks later, Salesforce paid $3.6 billion for it.
The AI agent was nearing $100 million in ARR and growing at 3.5 times. It handles over two million customer conversations a week. But the interesting part is not the growth rate. It is the model underneath. Fin built Apex, a domain-specific model post-trained on an undisclosed open-weights base, that achieves a 73.1 percent resolution rate on customer service queries. GPT-5.4 and Claude Opus 4.5 both hit 71.1 percent on the same benchmark. Claude Sonnet 4.6 scored 69.6 percent.
A two-percentage-point edge does not sound like much until you consider that the gap between successive generations of frontier models is often smaller. Fin did not build a better general model. They took a commodity base and post-trained it on millions of real customer service conversations until it understood the specific patterns of how people ask for help and how agents resolve those requests. The model also responds in 3.7 seconds and runs at roughly one-fifth the cost of calling frontier APIs directly.
Salesforce already had Agentforce, its own enterprise AI agent platform. What it lacked was a model that had been sharpened on actual customer conversations at scale, and a customer base already using it. Fin brought both. The acquisition is not about buying intelligence. It is about buying specificity.
The rename is the tell. When a company changes its name from its platform to its AI feature, it is acknowledging that the feature ate the platform. Intercom was a suite of tools for customer communication: live chat, help centers, product tours, email campaigns. Fin is an AI agent that resolves customer problems. The suite still exists, but it is infrastructure for the agent, not the other way around.
This is likely the first major acquisition where the buyer paid a premium specifically for a vertical AI model built on open weights. The pattern will repeat. General-purpose frontier models are commodity inputs. The value accrues to whoever fine-tunes them on proprietary data in a specific domain and then proves, at production scale, that the result outperforms the general version. Fin proved it in customer service. The next Fin will prove it in legal, medical, or financial workflows.
The rename was not marketing. It was an honest description of where the value had migrated. Salesforce understood that, and paid accordingly.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)