AI Outbound Sales Agent

An outbound calling agent that supports the sales workflow end to end. Before a call it pulls up what we already know about the customer; during the call a voice agent handles the conversation; afterwards it summarises what was said, records the outcome in Zendesk, and triggers the follow-up ticket for a human to review.

The point was never to replace the call. It was to remove the twenty minutes of lookup and write-up wrapped around it.

Making it sound less like a robot

The interesting problem here wasn’t the automation — n8n and webhooks handle that. It was that the first working version was technically fine and completely unbearable to talk to.

Real people interrupt. They pause mid-sentence to think. They say “sorry, what?” and expect you to back up rather than restart. An agent that waits for a clean silence before responding feels like a phone tree, no matter how good the voice sounds. Most of the work went into the timing — when to let a pause sit, when to acknowledge without taking over the turn, and how quickly to hand off to a human when the conversation clearly wanted one.

The lesson I keep coming back to: the voice quality was the easy part. The conversational behaviour around it is what decides whether someone stays on the call.