Intercom’s marketing for Fin is good. Maybe too good. “Resolves 67% of support queries autonomously” is a compelling number, and it is not fiction — we have seen clients hit that and occasionally exceed it.
But getting there from a fresh implementation is not what the landing page implies. Here is what it actually looks like, based on several implementations we have done over the past year.
What Fin genuinely does well
For queries that have a clear answer in your knowledge base, Fin is fast and reliable. Password resets, billing questions with standard answers, integration setup steps, refund policy — if it is documented clearly, Fin handles it well. Customers get answers in seconds. Agents stop seeing the same five questions forty times a day. This part of the pitch is real.
The routing intelligence is also better than I expected going in. Fin does not just fail gracefully when it cannot answer — it asks clarifying questions that make the escalation to a human more useful. By the time a ticket reaches a human agent, Fin has already gathered context that would have taken the agent two or three back-and-forth messages to collect. That is a genuine time saving even for queries Fin cannot resolve.
The 67% number is real but conditional
Here is what the marketing does not foreground: the resolution rate is a function of your knowledge base quality, not Fin’s capability.
In every implementation we have done, the first thing we do is an honest audit of the existing knowledge base. Almost universally, we find articles that are outdated, articles that contradict each other, articles that answer adjacent questions but not the actual question customers ask, and significant gaps where common queries have no documentation at all.
Companies that go live on a mediocre knowledge base see 25–35% resolution rates initially. Getting to 60%+ requires spending several weeks identifying what Fin is failing on, writing the missing content, and improving the structure of existing articles so Fin can retrieve from them more accurately.
This is not a criticism of Fin — the underlying model is solid. It is a note about expectations. The knowledge base work is not incidental. It is the majority of the implementation effort.
What a real implementation involves
The configuration side of a Fin implementation — connecting it to Intercom, pointing it at your knowledge base, setting up basic handoff rules — takes a few days. The actual work is everything else.
You need to design the intent mapping: which categories of query does Fin handle, which does it route, and under what conditions? You need to build the escalation flow logic so that human agents receive tickets with useful context rather than just a frustrated customer. You need to set up CSAT tracking specifically on Fin-handled conversations so you can monitor where it is underperforming. And you need to integrate Fin’s conversation data with whatever system you use to track customer issues.
We typically scope this at 3–5 weeks for a company with a reasonable existing knowledge base. Shorter if the content is genuinely good. Longer if it needs significant work.
Who should and should not buy Fin
Strong fit: SaaS products with consistent functionality, high support volume, and queries that are genuinely answerable with documentation. E-commerce with straightforward order-related queries.
Weaker fit: companies where most support queries are highly contextual, require account-specific investigation, or involve nuanced judgment calls that cannot be documented. Fin will escalate most of these correctly — but the resolution rate will be low and the ROI calculation changes.
The question to ask yourself before buying: what percentage of our current support volume consists of questions that have a definitive, documentable answer? If the honest number is above 50%, Fin is probably worth it. Below 30%, you are buying a sophisticated routing tool rather than an autonomous resolver.