Jesse Zhang - Building Decagon - [Invest Like the Best, EP.443]

What It Covers

Jesse Zhang, CEO of Decagon, explains building AI customer service agents that automate support conversations, achieving product-market fit through systematic customer discovery, competing in intense AI talent wars, and scaling enterprise deployments with 500-person call centers.

Key Questions Answered

• •Finding Product-Market Fit: Ask potential customers exact pricing they would pay for solutions during discovery calls, then probe deeper on approval processes, ROI calculations, and budget allocation. This systematic qualification process revealed customer service had 10x higher willingness to pay than other AI use cases explored.
• •Enterprise AI Deployment Strategy: Start with 5% of conversation volume, monitor resolution rates and customer satisfaction metrics for one week, then rapidly scale to full deployment within weeks. The built-in escalation path to human agents reduces risk and enables fast enterprise adoption compared to other AI use cases.
• •Voice-to-Voice Model Challenges: Direct voice-to-voice AI models have 8x higher hallucination rates than text-based approaches but offer superior latency and emotional understanding. Current enterprise solutions convert voice to text for accuracy checks, sacrificing some naturalness for reliability until voice models improve substantially.
• •Forward-Deployed Engineer Economics: The forward-deployed engineering model only works economically with clients paying minimum $1 million annually, not the $50,000 deals many startups pursue. Scaling requires building products that don't need dedicated engineers per customer, as hiring constraints prevent rapid growth otherwise.
• •AI Cost Margin Philosophy: Run zero or slightly positive gross margins on AI inference costs today rather than optimizing prematurely, because exponential improvements in model efficiency and declining compute costs will naturally improve economics. Focus engineering time on customer acquisition and product quality instead of margin optimization.