How AI Reversed the Order of Product Design

Toss's customer service chatbot suffered from a user drop-off rate of up to 60% due to its existing menu-navigation structure. To solve this, the team attempted to transition to an AI-based chatbot that understands users' natural language and grasps their intent.

Previously, design followed the order 'analysis - type organization - scenario writing - prototype,' but in this project, they adopted a method of first generating a draft scenario with AI and then immediately validating it with a prototype. In particular, instead of policy data, they trained the AI on actual consultation data to implement natural scenarios similar to how counselors actually respond.

For efficient validation, the following process was built:

• Scenario Hub: Built an environment that combines various user situations (whether payment was made, whether benefits were used, etc.) to instantly test conversations
• Rule-based Optimization: Instead of modifying individual scenarios, they created common rules such as 'prioritize resolution' and 'no guessing' and reflected them in the AI, simultaneously improving the quality of all scenarios at once

This reverse design (Experience-first) approach, by working backward to define the data and APIs needed for implementation, demonstrated high productivity, completing everything from planning to enhancement in just 3 weeks.