Resolution Bot lessons for support AI
How AI is transforming B2B SaaS
Resolution Bot showed that Intercom had already learned the hardest lesson in support AI years before the LLM wave, accuracy alone is not enough, the setup burden and ROI have to work too. Resolution Bot could reach roughly 50% resolution rates, but only when support teams spent real time writing answers, mapping intents, and maintaining rules. The fact that it still produces millions in revenue means that for customers with repetitive, well documented questions, that manual tuning still solved a real pain point and created a bridge into Fin.
-
Resolution Bot sat in the second chatbot era between phone tree bots and LLM agents. Instead of forcing users to click through menus, it tried to infer intent from typed questions and return a prewritten answer. That was a meaningful product step, but still depended on humans curating the answer library.
-
This also explains why Intercom moved so fast after ChatGPT. The company already had an ML team, historical bot data, answer libraries, and a clear view into where older bots broke down, hallucination risk, brittle rules, and too much customer setup. Fin was not a cold start, it was the third generation built on top of that operating knowledge.
-
The commercial signal matters. Intercom grew from about $275M of revenue in 2023 to about $343M in 2024 as AI products helped reaccelerate growth, while the broader market shifted from per seat help desk pricing toward per resolution pricing. Resolution Bot was an early proof that support software could charge for solved outcomes, not just agent seats.
Going forward, the winning support vendors will keep combining the old strengths of products like Resolution Bot, tight control, curated answers, and workflow rules, with LLMs that can handle open ended conversations. That points to a market where the best products are not simple model wrappers, but full support systems that connect docs, inbox, customer data, and automated actions in one loop.