Twelve years ago our four-person studio in Tokyo built feather for Twitter, Hit #1 in Japan’s App Store. When the Twitter API changes killed third-party clients, our revenue died overnight. 2025-01: we locked ourselves in our office and decided, “Let’s go B2B; AI can turn raw logs into stories, surely sales teams will love that.” We shipped an MVP of Eyes Mail: GPT-4 generated subject-line variants, measured open/CTR, promised auto-optimization. Reality: customers pasted their CSV, clicked “Send all”, and ignored everything else. One founder spent a week perfecting an AI-guided wizard that got *zero* clicks. The painful aha-moment: repetitive bulk mail is a chore; AI shines only where humans care about nuance. We were adding latency, not value. So we deleted 70% of the code and rebuilt the product as a single clean page: upload list → write message → send → get an instant plain-text report. Tasks that took minutes now take ~30 s. With that friction gone, customers finally started asking deeper questions: “Which segments convert?”, “Can your AI draft follow-ups?” That’s now Eyes Analytics, a separate layer that unifies mail + web data and, *only when you ask*, spins up AI agents to dig for answers. Lessons we paid for: 1. “Ask if you want KPIs” is useless—everyone says yes. Instead, watch what they do when every extra click hurts. 2. A half-baked AI feature can be worse than none; make the boring path blazingly fast first. Happy to answer anything, tech stack (React / TailwindCSS / AWS Bedrock / Amplify), metrics, Japan-specific SaaS hurdles, etc.

Eyes Mail

Send thousands of emails from Gmail. Dead simple.

Send thousands of emails from Gmail. Dead simple.