ENTERPRISE SAAS • B2B • AI
Leading the zero-to-one design of a real-time AI guidance layer that surfaces live coaching and sentiment signals without disrupting agent focus during high-stakes customer interactions, built to live natively within Salesforce.
NICE AI Copilot for Agents
ROLE
Senior UX Designer
TEAM
1 Design Manager, 2 UXR, 1 Jr. Designer, 2 PMs
PLATFORM
B2B Enterprise SaaS, Agent Desktop
SCOPE
Product design, UX research, design systems, AI tooling
OVERVIEW
In enterprise contact centers, manual Quality Assurance (QA) only evaluates 1–3% of interactions, leaving agents without support when they need it most, during the call.
To solve this, I led the 0 to 1 design of an AI Copilot (Real-Time Interaction Guidance) tasked with surfacing live coaching, sentiment analysis, and next-best-action prompts directly to agents.
The core architectural challenge wasn't algorithmic capability, it was human cognitive bandwidth.
Determining what mattered, when it should appear, and how to present it. Misjudging the interface meant the AI would either be ignored completely or become a dangerous distraction during high-stress interactions. This dynamic guidance needed to live natively inside the Salesforce CRM, rigidly governed by its own design language (Lightning Design System) and constraints.
+ 80 of the Fortune 100
CONSTRAINTS & APPROACH
The process required navigating the tension between proactive AI intervention and a seamless CRM workflow.
Fun fact: COVID happened during this and we transitioned to WFH.
PROCESS
Conducted thorough on-site and remote user research, combining quantitative data with moderated usability testing to analyze real agent behavior under live pressure.
Collaborated directly with Salesforce product managers to map the interface with the Lightning Design System, ensuring the AI components felt natively integrated into the environment agents already trusted.
Defined the logic and interaction patterns for AI interventions, establishing strict rules for timing and visual hierarchy to manage cognitive load.
Evolved the broader CXone design system and component library, enforcing WCAG accessibility at enterprise scale while mentoring 2 junior designers on complex interaction standards.
Iconography & Color were both used to communicate Sentiment at a glance, and required the reason why.
Knowledge Base Integration provided custom guidance.
OLD FLOW: Navigate to a different tab, click search, type keywords, and find the right information.
NEW FLOW: AI Copilot generates helpful information with interaction context live with actions to Save and Share already available.
KEY INSIGHTS
!
Contextual restraint over data density
Just because AI detects a sentiment shift does not mean it warrants an immediate UI prompt. Guidance had to be strictly actionable moments to prevent alert fatigue and user burnout.
SOLUTION: Sentiment and Guidance can be collapsed by the user, and still surfaces key info.
!
Designing for the "glance"
Agents cannot afford to take their eyes off primary customer data. UI required strict visual hierarchy so guidance could be understood in milliseconds without breaking conversational flow.
SOLUTION: Systematic informative iconography, color palette, and language.
!
Ecosystem adherence builds trust
To drive rapid adoption, AI Copilot could not look like an intrusive third-party widget. It had to behave as a seamless, native extension of the Salesforce ecosystem.
SOLUTION: Expert understanding of Lightning Design System and integration partnering with Salesforce PMs.
THE PRODUCT
AI Copilot: Simplify agent work, amplify human impact
The resulting AI guidance layer reads context, contact sentiment, conversation history, and knowledge base data in real-time to surface coaching cues directly inside the agent's existing workflow.
It is live, shipped, and you can check out the interactive product demo here.
A live phone call
An omnichannel interaction
Impact & Business Outcomes
The deployment across NICE's enterprise client base drove massive operational efficiency and financial efficiency:
100%
Interaction QA
From a manual 1–3% baseline to total operational oversight
23%
Drop in Agent Attrition
Eliminating $2.1M–$5.25M in annual replacement costs for a 500-seat contact center
55%
Reduction in AHT
Freeing agent capacity to handle more interactions and upsells
*Average Handle Time
REFLECTIONS
Leading this zero-to-one AI integration reinforced that successful AI product architecture is rarely about maximizing machine output, it is about respecting human bandwidth.
Designing an assistant that dynamically intervenes during live interactions required restraint and an understanding of cognitive limits under pressure.
Navigating the rigid constraints of the Salesforce Lightning Design System taught me how to embed complex, predictive logic into an established enterprise ecosystems natively.
This structural mindset is what I carry forward into building complex zero-to-one platforms and human-centered AI experiences today.
