The single metric we tracked that changed how we build AI interfaces

Hey Product Hunt community!

While building multi-model software over the past few months, our team fell into a common trap: focusing almost entirely on response speed and model capability.

However, after looking closely at user session recordings, we noticed something unexpected:

Users weren't abandoning chats because an AI model gave a bad answer. They were quitting because they spent over 40% of their total session time just setting up context, pasting background data, and switching tabs.

This changed our whole product philosophy. We realized that solving UX friction and context retention creates far higher user satisfaction than just upgrading to a newer LLM.

As builders and product managers:

• What is one non-obvious metric or user behavior that completely changed how you design your product?

• How do you balance adding flashy new features versus polishing micro-frictions?

Would love to learn from your team's experience!