Decagon Became a Unicorn by Simply Listening to Customers

“You Just Have to Listen to Customers” For Jesse Zhang and Ashwin Sreenivas, when they founded Decagon in 2023, the two had seen the same pattern play out across companies. Customer experience systems were rigid, resource-intensive, and no longer fit for the way businesses operate. Despite years of investment, customer support remained one of the least efficient functions in the enterprise stack. The underlying problem was clear. In an exclusive conversation with AIM Media House, Jesse Zhang laid out the disconnect he saw in the market. “Customer experience or customer support really is a very exciting category,” he said. When customer experience is asked about by an average person, seven out of ten responses would likely be negative and only three positive. Zhang had seen those inefficiencies up close. His first company was a consumer app. A lot of customer support that was just done manually by me. All the systems had to be built by him manually. At the time, GenAI models weren’t widely available, so older forms of automation were relied upon. Through that process, empathy was developed, both for the support operation and the tooling required to make it successful. That experience, combined with the arrival of GPT-4, shaped what Decagon would eventually become, an AI support platform built for speed, control, and end-to-end automation. “Simply put, we're an AI customer service agent,” Zhang said. Their whole vision is that in the future, any user of any product, any time is going to have a personal concierge there that they can talk to, respond instantly, and have access to all their data. In under two years, Decagon has grown into a billion-dollar business or more better to say a Unicorn. In June 2025, it raised a $131 million Series C round at a $1.5 billion valuation, bringing total funding to $231 million. The round was co-led by Accel and Andreessen Horowitz, with participation from Bain, A*, Bond, Forerunner, Avra, and Ribbit Capital. Its AI agents now handle millions of customer interactions per month for enterprise clients across industries including Substack, Oura, Duolingo, and Bilt. Building Around the Signal In the early days of Decagon, Jesse Zhang and Ashwin Sreenivas didn’t start by writing code but they started by listening. The duo spent their first weeks in conversation with large enterprise stakeholders across industries, testing assumptions and collecting early signals. “Let’s be very open-minded,” Zhang recalled. “Let’s be very attuned to the signals that we're getting from real customers.” Dozens of conversations followed, and a pattern quickly emerged that AI had broad potential, but customer support stood out. It wasn’t just a viable use case but the most obvious one. “The ROI was clear. The scale was massive,” Zhang said. “It made sense.” The pair also found that while many in the market believed GenAI had promise in CX, the tools already available weren’t delivering on that potential. “At the time, we were just two people. No one knew about us,” Zhang said. And even then, they were getting serious interest. That was a signal that the existing solutions weren’t good enough. Breaking from the SaaS Mold Decagon’s early product strategy took a deliberate turn away from how most SaaS companies approached AI. “A lot of the traditional players are building AI solutions the same way they built their SaaS tools,” Zhang said. That, in his view, was the problem. He pointed to Salesforce as a classic example while acknowledging being a very successful business. “But imagine I need a car. If I go to Salesforce, they’ll build me a car, but first, it’ll be built from scratch, and second, it’ll be built their way. You spend months of time, money, and specialized effort to get it up and running.” Even once operational, the platform remained opaque. A car is received, but it can’t be driven easily because it has been built entirely using Salesforce’s custom configuration. And when something breaks, it must be returned to the vendor to be serviced, treated like a black box that only they are able to access. The AOP Framework Decagon’s counter-approach was built around what it calls Agent Operating Procedures (AOPs), a structured framework designed to mimic how human teams operate using SOPs. Unlike complex configuration interfaces or rules engines, AOPs can be written and updated in natural language. “The idea,” Zhang explained, “is to empower business users, not engineers to define how the AI behaves.” The AOP model enables real-time updates, clearer governance, and embedded transparency. Decagon’s architecture includes five layers: the AI agent, a routing system to escalate to humans when needed, a QA system known as Watchtower, a layer for conversation-level analytics, and a recommendation engine that flags gaps or suggests new AOPs based on ticket patterns. “If the system doesn’t know something, it should be able to recognize that,” Zhang said. “It should know, ‘Hey, I can’t help you with this—

Decagon Became a Unicorn by Simply Listening to Customers

Keep reading. It's free.

Related stories