Cheng emphasizes the 'explore-exploit' framework: constantly testing new ideas while scaling proven ones. At Grammarly, showing premium features to free users doubled conversions, proving that reverse trials outperform time-limited ones. Strong retention—30-40% D1—is critical for subscription apps, but mature products grow most from reactivating lapsed users. Chess.com runs nearly 1,000 experiments yearly, using AI to accelerate ideation and analysis. Gamification works when it combines core loops, metagames, and social profiles, as seen in Duolingo’s success. High-agency hires often beat experienced ones because they act fast and learn quickly. Cheng prefers mid-sized companies where impact and scale balance. Past failures, like Chariot Direct, taught him to define the real problem before building. Ultimately, sustainable growth comes from deep user understanding, relentless iteration, and integrity in decision-making.