Proactive

A note

This project is under NDA—I can only share the approach and thinking, not the specific deliverables. What I can say is that the most valuable moment wasn't a polished deck. It was when a stakeholder reached for the controls mid-meeting and said "wait, what if we changed this?" That shift—from critique to co-creation—is what this prototype was built for.

A prototype for prototyping

This is a prototype that prototypes. The thing being designed was a proactive chatbot—an AI that surfaces on a page without being asked. But rather than committing to one version of that behavior, the prototype was built to expose the design variables live, so the team could debate and tune them with stakeholders in real time.

The core question: what defines the right moment for an AI to speak up? That question has no single answer—it depends on timing, context, and how much the product trusts its read of the user. The prototype makes those dependencies visible and adjustable.

The demo

Choose a trigger type, set its parameter, then step through the three states: page loads, trigger fires, chat opens. The right panel shows exactly how the chatbot would appear on the page.

Variables as questions

Each trigger type encodes a different belief about when AI should interrupt:

• Behavioral — How many signals are enough? At 2 views the AI feels presumptuous; at 5 it may have missed its moment. The threshold is a decision about trust.
• Idle — How long is too long to wait? 5 seconds feels intrusive; 30 might lose the user. The window defines the AI's sense of urgency.
• Contextual — How confident does the system need to be? Low confidence risks noise; high confidence risks silence. The threshold shapes the AI's personality.

The takeaway

Prototypes that invite tweaking surface what people actually care about. The real design work happened not in building the chatbot, but in the conversations this prototype made possible—about timing, tone, and when an AI earns the right to speak first.