Jesse Zhang - Building Decagon - [Invest Like the Best, EP.443]
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
Jesse Zhang, co-founder and CEO of Decagon, shares his journey building one of the fastest-growing AI customer service companies. (05:00) Jesse discusses his systematic approach to finding product-market fit by asking potential customers exactly how much they'd pay for solutions, ultimately discovering that customer service emerged as the highest-value use case. The conversation explores the intense competitive dynamics of building AI companies today, why customer service has become one of the clearest AI enterprise applications alongside coding, and how Decagon's approach to deploying conversational AI agents at scale differs from traditional chatbots. (24:20) Jesse also discusses strategic decisions around proprietary model development, the future of voice-to-voice interactions, and the cultural elements that drive high-performing AI teams.
- Core themes include systematic customer discovery processes, competitive AI company building, enterprise AI deployment challenges, and the evolution from customer service automation to comprehensive brand interfaces
Speakers
Jesse Zhang
Jesse Zhang is the co-founder and CEO of Decagon, one of the fastest-growing AI customer service companies. Before founding Decagon, Jesse started and successfully exited a company called Low Key, which built high-performance video capture software for video games, exiting in 2021. Jesse has a strong background in competitive mathematics, having participated in Math Olympiads and attending camps for top performers during high school. He graduated college a year early to pursue entrepreneurship and brings a systematic, problem-solving approach to company building that stems from his competitive math background.
Patrick O'Shaughnessy
Patrick O'Shaughnessy is the CEO of Positive Sum and host of the "Invest Like the Best" podcast. He conducts in-depth conversations with successful founders, investors, and business leaders, focusing on extracting actionable insights for ambitious professionals. Patrick has built a reputation for thoughtful interviews that explore both strategic business decisions and the personal philosophies that drive high-achievers.
Key Takeaways
Systematize Customer Discovery with Direct Pricing Questions
Jesse developed a systematic approach to validating business ideas by going beyond typical customer interviews to ask specific pricing questions. (14:37) Rather than simply asking if customers want a solution, he would dig deeper: "If we built this for you, exactly how much would you pay for it? Would your boss need to approve it? How would you present ROI to leadership?" This forcing function helped differentiate between genuine demand and polite interest. When comparing multiple use cases, customer service consistently generated responses in the low-to-mid six figures, while other ideas struggled to justify even basic subscription costs. This systematic qualification process revealed that most "exciting" ideas weren't commercially viable, but customer service represented genuine urgent need with clear economic justification.
Focus on ROI Clarity and Built-in Escalation Paths for Enterprise AI
Customer service emerged as an ideal AI use case because it combines easy-to-quantify ROI with natural risk mitigation. (21:59) Companies already track conversation volume and resolution rates, making it simple to calculate savings when AI increases resolution from 15-20% to 50-80%. Unlike other AI applications where failure modes are unclear, customer service has built-in escalation paths - if the AI agent can't handle something, it simply transfers to a human using existing call center infrastructure. This gives enterprise customers comfort to start small with 5% of volume and scale rapidly when performance meets expectations, often expanding to full deployment within weeks once metrics confirm success.
Align on Success Metrics Before Building AI Agents
The biggest challenge in deploying enterprise AI agents isn't technical capability but defining "what good looks like." (26:41) Jesse discovered that companies rarely have unified answers for how conversations should be handled across their complex organizations. Success requires creating comprehensive evaluation suites with thousands of test conversations that establish clear standards for tone, brand guidelines, and appropriate responses. This process forces different stakeholders to align on expectations and provides quantifiable metrics for AI performance. Without this upfront alignment, even technically impressive AI systems fail to gain enterprise adoption because there's no clear definition of success.
Design for the Spectrum Between Flexibility and Control
Enterprise AI agents must operate along a spectrum from maximum flexibility to strict compliance depending on the use case. (31:17) Traditional customer service systems used rigid decision trees that frustrated customers when their needs didn't fit predefined branches. LLMs unlock natural conversation flow, but enterprises need the ability to enforce strict guardrails for regulated use cases while allowing flexibility for basic inquiries. Decagon's success comes from building systems that can be configured anywhere along this spectrum - from free-form conversation for account questions to rigid step-by-step processes for compliance scenarios. This adaptability is crucial because different use cases within the same company may require vastly different approaches.
Prioritize Customer Experience Differentiation Over Technical Complexity
The biggest surprise in AI customer service wasn't technical capability but how dramatically customer behavior changes when the experience feels genuinely different. (32:52) Before implementing AI, one in three customers would immediately demand to speak with a human agent, having lost trust in automated systems. After implementing well-designed AI with clear upfront communication about the improved experience, this dropped to one in twenty. The key insight is that customers are willing to engage with AI when they believe it will actually help, rather than waste their time. This suggests that success in AI applications often depends more on user experience design and trust-building than on raw technical sophistication.
Statistics & Facts
- When Jesse was systematically evaluating different AI use cases through customer interviews, customer service consistently generated willingness-to-pay responses in the low-to-mid six figures, while other ideas were an order of magnitude lower. (20:25)
- Voice-to-voice AI models currently have approximately 8x higher hallucination rates compared to text-based models, making them challenging for enterprise deployment despite their natural interaction benefits. (37:47)
- At Oura Ring, one of Decagon's customers, the percentage of users who immediately demanded to speak with a human agent (saying "agent, agent, agent") dropped from one in three to one in twenty after implementing their AI system. (32:58)
