[Yotpo] Allow external agents to access data in real-time

Transforming a SaaS Company into AI-Native is tough I’ve been talking to a lot of SaaS founders recently—companies like Yotpo, all between $30M–$300M in ARR.Yeah sure, everyone wants to be Cursor, Lovable and the rest, but 99% of companies today are not. Everyone agrees: We need to transform our companies to become AI-native. Everyone understands: We’re at a disadvantage. We’re carrying hundreds of thousands of lines of code written in a pre-AI world, With teams that weren’t born into this shift. But the desire to lead is real. Tobi’s memo sparked something across the board(go read it). Still, many of us are asking: what does “AI-native” actually mean operationally? This isn’t about putting AI into the product. It’s about using AI to transform how we work—how fast we move, how lean we operate, how productive we can be. So I thought I’d share what we’ve actually been doing at Yotpo. Not theory. Practice. 🧠 18 months ago we made our first real move. We created a dedicated internal automation team. $500K budget. Clear mandate: use AI to save costs. Looking back now—here’s what I’d do differently: 1. We focused too much on cost savings. That was our OKR. Cut costs. But cost savings are finite. Throughput, speed, productivity? That’s exponential. If I had to start again, I'd optimize for output, not just efficiency. 2. We lacked ruthless prioritization. There were a thousand shiny use cases. The team bounced around too much. What we learned: early momentum matters more than big bets. Start with small wins. Ship fast. Earn trust. 3. We underestimated how critical good data is. We thought the model would do the work. But garbage in, garbage out. Support was the first domain we tackled. It was mature, and measurable. We didn’t build our own LLM—we built an architecture that can switch models easily. Today in our leading product: 100% of our support tickets go through AI 50%+ are deflected CSAT stayed stable (took a few months of iteration to get to) But it wasn’t magic. We had to fix our documentation. So we built an AI to identify knowledge gaps and rewrote dozens of help articles. Only then did the deflection rate jump. 📌 Real takeaway? The AI-native journey isn’t plug-and-play. It’s a series of systems improvements. It’s iteration. It’s architecture. It’s internal conviction and patience. Next I will share about marketing and engineering —what’s working, what’s not. But if you’re a SaaS operator going through this: → What internal use cases have actually moved the needle for you? → What would you do differently if you had to start today? This is the biggest technological opportunity—and risk—of our careers.