The challenge
Stackline's five-person support team was handling 400+ tickets per week — mostly the same questions about billing, account access, integrations, and onboarding. Response times stretched to 6–8 hours on busy days. Agents were copying answers from a messy Notion doc, and customers were opening duplicate tickets when they didn't hear back fast enough.
What we built
Stackline already ran support through Intercom — we didn't set that up. We built a custom AI assistant that plugs into their existing inbox, product docs, and Notion knowledge base. Claude reads each incoming message, checks the customer's account context, and either resolves the request directly — password resets, plan questions, integration setup steps — or drafts a reply for human review. An n8n workflow syncs resolved answers back into Notion so the knowledge base stays current. Complex or sensitive cases escalate to the right agent with a full conversation summary and suggested next steps.
The outcome
Within the first month, 68% of inbound tickets were handled end-to-end by the assistant without human involvement. First response time dropped from hours to under 30 seconds. Support agents now focus on bugs, edge cases, and high-value accounts — and customer satisfaction scores went up because people got answers when they actually needed them.
The before state
Every ticket landed in the same queue. An agent would read it, search Notion or Slack for the right answer, paste something together, and move on. Billing questions, API key issues, and 'how do I connect X?' requests ate most of the day. After-hours and weekend tickets sat until Monday morning.
What we built
The AI layer sits on top of their existing help desk — not a replacement for it. It pulls from a structured knowledge base of product docs, FAQs, and past resolved tickets, and reads account-specific context like plan tier, connected integrations, and recent activity. When confidence is high, it responds directly in the inbox. When it's not, it flags the ticket and pre-writes a reply for an agent to approve or edit. Every new resolution updates the knowledge base automatically.
How it runs today
The support team starts each day with a queue of only the tickets that need human judgment — technical bugs, refund requests, and enterprise accounts. Everything else is already answered. Agents review AI-handled conversations weekly to catch gaps, and the assistant gets sharper over time. Stackline hasn't needed to hire additional support headcount despite 40% user growth.