LLM Customer Journey
2026
#UX & Interaction Design. #AI Training #Motion UI
Overview
To enhance sales training, we introduced AI-powered virtual customers that allow salespeople to practice product conversations in a realistic, dynamic environment. The system evaluates whether users:
✔️ Hit key product talking points
✔️ Deliver a persuasive pitch
✔️ Successfully guide the conversation to a sale
By simulating different customer personas and scenarios, this tool helps salespeople refine their approach in a natural, LLM-driven dialogue.
Step 1: Reducing Complexity in Scenario Setup
Context
Since LLMs can simulate a wide range of customer interactions, we needed to narrow down conversation topics for more structured training. However, this required users to manually define multiple parameters before starting a session:
Early stage mockup shows mutiple stepes are required before the training sesstion
Select a product category➡️ Choose a specific model➡️ Define a customer scenario➡️ Generate a customer persona
Design Solution:
💡 A Visually Intuitive “Bubble Flow” Interaction
I designed a progressive selection experience that transforms the setup process into a playful, visual journey:
Highres mockup of “Bubble Flow” interaction
Impact & Key Takeaways
✅ Reduced cognitive load → A one-page, immersive interface replaces traditional multi-step forms.
✅ Increased engagement → A motion-enhanced, evolving UI makes setup feel more intuitive and enjoyable.
✅ Better user adoption → Salespeople feel in control of the customization process, making training more approachable.
Step 2: Reducing Perceived Latency in Digital Human Conversations
Context
In a voice-based digital human interaction, users are highly sensitive to response timing and synchronization between text and speech.
Before (Original Experience)
Problem
🕖 LLM text response latency: ~1.7s after user speech ends,text-to-speech with lip-sync: additional ~2.0s
🕖 Text appeared first, audio followed later,the gap created a persistent sense of delay, despite acceptable total latency
Design Solution
💡Introduced transitional animations to absorb system processing time,used animation to communicate “thinking” and manage user expectations
💡Deliberately delayed text display to align with audio readiness
💡 Adjusted animations for both direct send and edit-then-send flows to ensure consistent perceived waiting time across branches
After – Direct Send Flow
After – Edit-Then-Send Flow
Outcome
✅ Unified perceived latency across multiple interaction paths
✅ Improved conversational flow and coherence
✅ Interaction felt more natural and human-like