Agents Update Docs to Train Bots
Eoghan McCabe & Des Traynor, CEO and CSO of Intercom, on the AI transformation of customer service
This is the product wedge that turns an AI bot from a cheap add on into the system that runs support. When the same agent workspace also controls bot training and the help center, every escalated ticket can become a permanent fix. A rep answers the customer once, updates the source article, and improves the bot for the next identical question. That creates a compounding workflow advantage over stitched together stacks where the inbox, bot, and docs live in separate tools.
-
Intercom frames support as AI plus humans, not AI instead of humans. In practice, Fin hands uncertain cases to a rep, the rep edits or approves the answer, and that interaction becomes training data and documentation. The support team shifts from repeatedly answering known questions to handling edge cases and teaching the system.
-
This is the core difference between an AI platform and a GPT wrapper. The hard part is not generating text, it is connecting the model to customer records, prior conversations, knowledge base articles, routing rules, and reporting, then letting agents write back into that system from one screen.
-
The comparison point is Zendesk and other modular help desk stacks. Zendesk centralizes tickets across channels and offers chat, voice, and knowledge base tools, but Intercom is pushing harder on one joined workflow where automated resolution, human takeover, and content updates feed each other. That makes resolution quality and product adoption improve together.
The market is heading toward platforms where every support interaction also trains the next one. As AI resolves more repetitive tickets, the winning vendors will be the ones that make human escalations productive instead of expensive. That favors vendors that own the inbox, the bot, the knowledge base, and the customer context in one loop.