Intercom’s AI reads your docs. It doesn’t notice the ones you never wrote.
Fin is a strong retrieval agent on top of a strong messenger. The question is what happens to everything it can’t answer — and whether anyone ever turns those into articles.
What Intercom is genuinely good at
Intercom has spent a decade on the messenger itself, and it shows: the inbox is fast, the routing is mature, and the reporting is the deepest of the three. If your support operation is large enough to need workflow builders and SLA rules, they have built all of that and we have not.
Fin is also genuinely good at answering from a well-maintained knowledge base. On a library that is already thorough, the difference between their retrieval and ours is not the thing worth arguing about.
Where we differ
Fin’s coverage gaps surface as analytics you have to go and read. In HeyThula the handover itself is the artifact: the customer’s phrasing, the human answer that followed, and a stamp saying no article covered it — sitting in a backlog that reads like a to-do list, because that is what it is.
Intercom expects the article to already exist. We assume the reason it doesn’t is that writing it is a project nobody has time for, so you do the task once with the recorder on and get a draft with steps and screenshots. That is the step that actually unblocks a small team.
Intercom charges per seat and then per Fin resolution on top, which quietly discourages putting more of your team in the tool. We charge for resolutions only, invite everyone, and don’t bill for the conversations the AI handed over.
If you need omnichannel routing, custom workflow automation, SLA management, or you’re running a support org with tiers and escalation paths, Intercom is the more complete operations product and switching would cost you capability. We’re aimed at the team where support is one person’s afternoon, not a department.