Third-Gen Chatbots Eliminate Setup
Intercom's $250M/year AI bet
This is what turned AI support from a niche feature into a real budget line. Earlier bots only worked after a team spent weeks writing answers, mapping intents, and tuning rules for the top few ticket types. Fin changed the setup burden by reading an existing help center, handling open ended questions in normal language, and falling back to humans when confidence is low, which made automation usable for the broad middle of SaaS companies, not just support teams with the scale to hand craft bot logic.
-
Resolution Bot could hit about 50% resolution, but only when customers invested heavily in programming and maintenance. That made it work more like a consulting project than a plug in product, which limited adoption even when the raw resolution rate looked good.
-
Third generation bots changed the workflow from rules authoring to document ingestion. Instead of staff pre writing every answer path, the model reads public docs, interprets the question semantically, drafts or sends the answer, and escalates edge cases to an agent inside the same support flow.
-
That lower setup cost also changed pricing and competition. Once a bot can deliver useful answers quickly, vendors can charge per resolved ticket, and the battle shifts to who has the best orchestration, reporting, integrations, and handoff between AI and human agents. That is why Fin, Sierra, and Decagon all look more like AI agents than old chat widgets.
The next step is from answering questions to completing work. The winners in support AI will not just read docs well, they will verify identity, check account state, take actions like refunds or plan changes, and do it across chat and voice, which pushes the market further away from scripted bots and toward full AI operated service desks.