Challenges

Fragmented channels, repetitive tasks, and growing technical demands limited the ability to scale support effectively

AssemblyAI provides advanced speech AI APIs to help teams build and integrate voice capabilities into their products. As the company grew, its customer-facing teams faced the challenge of supporting highly technical users—engineers, product managers, and designers—across multiple channels and workflows.

Internally, Support and Customer Success operated on fragmented tools: Slack, email, live chat, and custom alerting systems. Tracking core metrics like First Response Time (FRT) required manual effort, and internal knowledge often lived in scattered docs or team members' heads. Support agents found themselves answering the same Level 1 questions repeatedly, limiting their ability to focus on complex, high-value work.

"Our customers are developers who expect quick, actionable support. We needed a way to meet them where they work without slowing down."

Lee Vaughn, Manager of Support Engineering at AssemblyAI

Solutions

AI Agents and Runbooks automated responses and scaled technical support

AssemblyAI adopted Pylon to unify its customer communication channels and introduce automation into its support workflows. The team deployed an AI Agent, "Sonny", to handle incoming chat inquiries, powered by a robust public knowledge base.

By continuously improving Sonny’s capabilities through Pylon’s Runbooks, AssemblyAI expanded the range of inquiries the AI Agent could resolve. Runbooks allowed the team to automate responses for common but unpredictable behaviors, such as users dropping YouTube links without context or submitting unstructured questions.

"We started with an AI resolution rate in the high 20%. Now Sonny successfully resolves close to 50% of incoming chats. That has made a real difference in freeing up our team to focus on complex customer needs."

Lee Vaughn, Manager of Support Engineering

Runbooks made it easier to account for these edge cases and ensure that Sonny could route users to the correct resources or troubleshooting paths, significantly improving the customer experience without requiring human intervention.

"Runbooks have helped us handle weird edge cases much more intelligently. Instead of failing the conversation, the agent now guides customers to the right resources automatically."

Lee Vaughn, Manager of Support Engineering

Results

Faster response times, higher ticket deflection, and improved efficiency

With Pylon, AssemblyAI has scaled its support operations while improving responsiveness and reducing the burden on the support engineering team.

• Median first response time of 23 seconds for in-app chat support issues deflected by the AI agent, providing instant help to technical users without human intervention.

• 2x improvement in AI agent resolution accuracy, now successfully resolving ~50% of eligible incoming chats without human intervention.

• Automated coverage through weekends and off-hours, enabling AssemblyAI to maintain 24/7 chat availability without requiring full weekend staffing.

• Faster handling of edge cases through Runbooks, allowing the AI agent to manage complex but common scenarios more effectively.

"Pylon has helped us turn support into a system that improves itself. Every time Sonny answers a question, our team gets to focus more on higher-value work."

Lee Vaughn, Manager of Support Engineering