Salesforce Einstein
Conversational UX Design | Service Design | AI Interaction Design
TL;DR
I redesigned the Salesforce Einstein chatbot to feel more human and context-aware by improving tone, continuity, and first-time guidance through research-driven conversation design.
Date
Sep 2024 – Dec 2024
Role
Lead Product Designer
UX Researcher
UX Writer
Tools
Figma
Miro
Google Docs
Google Meet
Methods
User Interviews
Usability Testing
Introduction
As part of a UX initiative with Salesforce, we redesigned the Einstein chatbot to feel more human, context-aware, and brand-aligned. The goal was to move from static scripts to a conversational experience that adapts to users and maintains memory across sessions.
How might we design a human-centered AI chatbot experience for small businesses using the Salesforce website for the first time with specific focus on product discovery?
Problem
Chatbot Personality & Tone
Our goal was to develop a personality system that balanced professionalism with human-like empathy and engagement.
- Users preferred clear, structured, and jargon-free communication.
- Formal tone worked best for workplace tasks; empathetic tone was essential for troubleshooting; and fun tone was most engaging for product discovery.
- Final personality was modeled after a dolphin— adaptive, empathetic, intelligent, and user-friendly.
Context Aware Experience
Users experienced disjointed conversations because the chatbot couldn't remember past interactions, causing frustration and information loss across sessions.
- Conducted contextual inquiries using ChatGPT, Amazon Rufus, and Salesforce Einstein.
- Identified issues: poor thread continuity, rigid prompts, and lack of memory.
- Refined prompt patterns to support multi-session continuity.
Simplifying Product Discovery
How might we simplify product discovery and recall to support faster decision-making, as users expect quick, helpful info to guide exploration?
Optimizing Prompts for GenAI Interaction
How might we guide users to write better prompts for relevant, context-aware responses from a GenAI chatbot?
Chat Navigation Timeline
How might we help users quickly navigate long chatbot threads by jumping to relevant topics or past suggestions without scrolling?
Behavioral Questions For Clarifying Questions
User tests showed people prefer chatbots with natural, helpful conversation. How might we prompt users for clarification or elaboration without disrupting flow?
Re-Engaging Users During Inactivity
Salesforce values customer retention and aims to ensure users feel supported throughout the journey. How might we re-engage users during inactivity?
Seamless Agent Connection
Enabling a seamless connection to a sales rep when the chatbot fails to find the product the user needs, particularly when the user has a specific item in mind.
User Research through ChatGPT
We used ChatGPT as a benchmark for consistency across sessions, adaptability, and controllable tone-useful for testing personality, brand voice, and response patterns.
User Testing
We tested three personalities-formal, playful, and empathetic-with 15+ participants, grounded in a 16-type personality framework.
Key Findings
- Users preferred clear, structured, and jargon-free communication.
- Formal tone worked best for workplace tasks; empathetic tone was essential for troubleshooting; and fun tone was most engaging for product discovery.
- Final personality was modeled after a dolphin— adaptive, empathetic, intelligent, and user-friendly.
Iteration and Prototyping
We introduced a dolphin character to guide first-time users with an approachable, supportive tone. Research insights shaped UX writing to be clear, empathetic, and confidence-building from the first interaction.
What I Learned
This project taught me that UX isn't just about new features-it's about refining what exists to reduce friction. Thoughtful interaction design helps users navigate confidently and find what they need without feeling overwhelmed.