When Inbound Sells Itself
What changes when the first response to a prospect no longer needs a sales rep? Jeanne DeWitt Grosser, COO of Vercel, has built go-to-market teams at Google, Dialpad, and Stripe. In this episode of Office Hours, she joins Tomasz Tunguz to explain how Vercel put AI to work on inbound sales. In six weeks, an agent took on the work of nine of ten inbound SDRs. Conversion stayed steady, prospects received faster responses, and the team moved into higher-impact work.
In our conversation, we discuss:
- Why personalized outreach that failed in earlier attempts now works
- How one GTM engineer and a top SDR built the first version
- What six weeks of human review contributed before the agent ran independently
- Why a growing prompt gave way to 14 explicit rules
- Where the model makes judgments and where software enforces the process
- How a second agent monitors exceptions
- What happens to SDR careers when inbound qualification is automated
Three takeaways from this conversation:
1. Start with the expertise already inside the team. Vercel’s strongest SDR wrote the initial qualification prompt, while one engineer devoted roughly a fifth of his time to the project. Experienced reps reviewed the agent’s work before the team removed the human checkpoint. The starting point was a specific process and the people who understood it.
2. Separate rules from judgment. As the prompt expanded, the model did not consistently follow its instructions. Vercel moved fixed requirements into code. An existing Salesforce opportunity, for example, should trigger routing to its account executive. The model handles the decisions that require interpretation; a monitoring agent checks rule breaches.
3. Give the sales team a more valuable next job. Vercel promoted its inbound SDRs into outbound ahead of the usual timeline. Automation created room for more human conversations. With conversion maintained and responses arriving faster, the case for the agent extends beyond reducing the work required to manage inbound.
