When I first joined HubSpot’s Conversational Marketing team, the vast majority of our website chat traffic was managed by people. We had a global group of more than a hundred live sales agents — Inbound Success Coaches (ISCs) — qualifying leads, scheduling meetings, and routing conversations to the right sales reps. The system functioned, but it wasn’t built to scale. Every day, ISCs were handling thousands of chat messages from visitors looking for product details, asking support questions, or casually browsing. While we valued those interactions, they frequently distracted from high-intent prospects who were ready to talk to sales. We knew AI could help us operate more efficiently, but we weren’t interested in another rigid, scripted chatbot. We wanted something that could behave like a sales rep: qualify, guide, and sell in real time. That’s how SalesBot came to life — an AI-driven chat assistant that now manages most of HubSpot’s inbound chat volume, responding to thousands of questions, qualifying leads, booking meetings, and even directly selling our Starter-tier products. Here’s what we’ve learned in the process. How We Built SalesBot and What We Learned 1. Start with deflection, then scale to demand. When we initially rolled out SalesBot, our main objective was to deflect simple, low-intent questions (for example: “What’s a CRM?” or “How do I add a user to my account?”). We wanted to cut down on noise and give humans more time for nuanced, high-value conversations. We trained the bot on HubSpot’s knowledge base, product catalog, Academy content, and more. Today, we’re deflecting over 80% of chats across…