Case Study 03 · Dell IntelliAssist 2.0 · Dell Technologies

I rebuilt Dell's AI workspace into a full-screen sales assistant.

Evolving Dell's AI support from a cramped chat box into a full-screen workspace where 12+ APIs, source-tagged answers, and the sales portal's tools all sit under one conversation. Ticket volume fell 55%, self-serve adoption hit 92%, and reps finally trusted what the AI said.

Shipped · 2024 AI Tools Enterprise SaaS Sales Enablement GenAI

The Challenge

Dell employees were stuck inside a cramped chat window that surfaced shallow AI answers and hid the platform's real capabilities. Checking a single order meant opening four other tabs, then pinging a teammate anyway.

The Solution

IntelliAssist 2.0, a full-screen AI workspace that unifies 12+ APIs, source-tagged answers, and the centralized sales portal into one place. Fewer tickets, more self-serve, and trust in what the AI says.

55%

Ticket volume reduced

across enterprise support

92%

Self-serve adoption

on the new workspace

78%

Faster task completion

in testing, vs. 52% on legacy chat

IntelliAssist 2.0 landing screen showing the AI chat, prompt suggestions, and integrated sales tiles

IntelliAssist 2.0, the redesigned Dell AI sales workspace

Role

Senior Product Designerresearch, IA, interaction, visual

Team

Dell Technologies2 PMs, 8 engineers, 2 designers

Timeline

8 monthsdiscovery through pilot launch, 2024

Tools

Figma · Mural · UserTestingpilot telemetry, post-launch surveys

Note: names, workflow details, and screen interfaces have been modified to maintain Dell's confidentiality.

Snapshot

The short version.

If you only read three paragraphs, read these. Problem, change, and results in about a minute.

Problem

Dell employees worked inside a cramped chat box that surfaced shallow AI answers and hid the platform's real capabilities. Checking a single order meant opening four other tabs, then pinging a teammate. Adoption stalled and support tickets piled up.

Change

I rebuilt IntelliAssist into a full-screen AI workspace that unifies 12+ APIs, source-tagged answers, and the sales portal's tools under one conversation. Reps see where each answer came from, trust what they read, and stay inside one surface.

Results

Ticket volume dropped 55% across enterprise support. Self-serve adoption hit 92% on the new workspace. Task completion ran 78% faster vs. 52% on the legacy chat. 85% of pilot users reported easier access to critical tools.

Problem area

The chat box was tiny. The job it had to do wasn't.

Dell's enterprise sales teams needed answers about orders, configs, and customer history during live conversations. But they were stuck inside a cramped, outdated chat interface, toggling between four to six systems to assemble a single response. Tickets piled up at the support desk, AI adoption stalled, and reps quietly defaulted to bookmarks and pinging teammates. The goal was to unify scattered tools into one full-screen workspace that could replace the chat-plus-tabs workaround with a single source of truth.

What & Why

Tiny chat. Wasted AI. Scattered tools.

Three problems showed up at the same time. The chat surface was too small for real work. The AI underneath gave generic answers that no one trusted. And the tools that could have helped were spread across a portal nobody could navigate. The redesign had to fix all three.

💬

The What

A full-screen AI workspace that replaces the cramped chat with prompt suggestions, source-tagged answers, and the sales portal's tools sitting directly underneath the conversation.

🎯

The Why

The business needed a stable platform to integrate new AI capabilities and unify sales tools. Without it, employees faced disjointed processes, errors, and stalled enterprise-wide AI adoption.

Research

Low trust, low discovery, high workaround behavior.

Across user interviews, a clear pattern emerged. Employees lacked confidence in the tool, found it slow to navigate, and rarely discovered features on their own. The result was low adoption, fragmented workflows, and delays in customer resolution.

The legacy IntelliAssist AI Result panel showing an order details response with limited context and no follow-up paths

🧩

Incomplete answer context

The AI gave surface-level details without supporting context, related information, or follow-up questions. Users couldn't tell what to trust.

"The chatbot gives generic answers. I can't trust it." Sales Representative

Legacy IntelliAssist landing screen with a sparse search input that hid the AI's full capabilities

📭

Buried features, low adoption

The sparse landing screen failed to highlight the AI's full capabilities. Feature discovery was low, and employees defaulted to peer-dependent learning.

"I have to ping a teammate to get the right resource." Project Manager

Dell Technologies Sales Portal home with a wall of self-serve tiles that users routed around in favor of bookmarks

🔖

Workarounds replaced the tool

Search was so slow that users had quietly stopped using it. Bookmarks were faster, even when stale. The tool was being routed around, not used.

"I just use bookmarks. Searching takes too long." Data Engineer

Synthesis venn diagram, tiny chat plus wasted AI plus scattered tools converging into a unified workspace

Synthesis, three pain points converging into one unified workspace

Design Principles

Speed. Clarity. Trust.

I reframed the challenge around three principles. Every design decision, from role-specific prompts to source-tagged answers, was built to shift behavior, boost adoption, and reduce support dependency.

Principle 01

Speed

Help users get to the right answer faster
Cut the steps between question and decision
Pre-emptively suggest common queries

Principle 02

Clarity

Context-rich, human-readable responses
Scannable templates for repeating tasks
Visual hierarchy that surfaces what matters first

Principle 03

Trust

Sources attached to every AI response
Familiar fallback paths when AI falls short
Personalization that earns trust over time

Speed, Clarity, and Trust mapped to behavior change goals

🚨 Phased transition strategy

Self-serve tiles were prioritized at launch to ease adoption, with plans to phase them out as AI usage grows. Balancing familiarity with long-term efficiency, not picking one over the other.

Final Design

One workspace, five intentional decisions.

The shipped IntelliAssist 2.0 wraps a full-screen AI chat in a workspace that holds context. Underneath, five interface decisions made the new pattern work for employees who'd been burned by the old one.

Track 01

Unified workspace for seamless workflows

Annotated overview of the IntelliAssist 2.0 landing showing where the smart prompt suggestions (1) and familiar self-serve tiles (2) sit on the page

Overview, two coordinated moves on a single landing screen

Smart search prompt affordance

User testing revealed hesitation around what to ask. I introduced role-specific prompts to reduce friction, guide first-time use, and highlight AI capabilities. These soft nudges boost perceived relevance and support progressive discovery, all without overwhelming the interface.

Contextual Onboarding · Behavior Design

Smart search input with role-specific prompt chips visible at first load

Familiar cards carried over from the legacy interface

To ease adoption, I retained the familiar card layout from the legacy interface. This preserved mental-model continuity, reduced cognitive friction, and helped users feel more in control. A small but intentional decision that improved learnability and supported a smoother behavioral transition to the AI workflow.

Change Management · Mental Model

Full IntelliAssist 2.0 landing showing the search input with prompt suggestions above the self-serve tile grid

Track 02

Full-screen interface for contextual efficiency

Annotated overview of an IntelliAssist 2.0 result page showing the collapsible sidebar (1), follow-up prompts (2), and sources panel (3)

Overview, the full-screen workspace with sidebar, follow-ups, and sources in view

Collapsible resource sidebar

To give users a sense of control, I added a collapsible sidebar so they could self-navigate when the AI response fell short. It acts as an escape hatch, offering a familiar fallback path that boosted confidence and supported smoother adoption during the transition to AI-assisted workflows.

User Control · Flexibility · Fallback UX

Collapsible left-rail sidebar showing self-serve resource tiles like GenAI Sales Support and Order Tracking Hub

Follow-up and "search instead" prompts

These contextual suggestions help users reframe queries or explore related topics when AI responses fall short. They reduce blank states and reinforce conversational UX patterns, making the AI experience feel more responsive and guided.

Intent Refinement · Conversational UX

Follow-up question chips and a search-instead input shown below an AI response

Sources and related articles inline

I added visible sources and related links to increase the credibility of AI responses and help users feel confident in what they were seeing. This supports AI explainability, gives users a way to independently validate results, and builds trust without dependence, especially important during early AI adoption.

Transparency · AI Explainability · Trust Design

Right-rail panel showing the source for the AI response with related articles listed underneath

The Shipped Design

IntelliAssist 2.0, end to end.

The five decisions added up to one workspace. A full-screen AI surface that holds context, surfaces prompts, attaches sources, and keeps the legacy self-serve tools one click away. This is the shipped state, the same view a Dell sales rep sees on day one.

The shipped IntelliAssist 2.0 workspace at full width showing the unified AI chat, prompt rail, sources, and self-serve tiles in one view

01. Landing, the unified workspace at first open

IntelliAssist 2.0 answering How to gain access to ESP with steps, key resources, sources, related articles, and a search-instead area below

02. Knowledge query, ESP access with sources, related articles, and a search-instead fallback

IntelliAssist 2.0 answering an order status query with an AI suggestion card, expandable order details, timeline, shipment, and cancellation accordions

03. Order status query, structured response with expandable details and a direct source link

"This tool transformed how our teams operate. Finally, the right resources are at their fingertips, exactly when needed."

Director, Operational Intelligence · IntelliAssist 2.0 launch retrospective

Impact

Three measurable shifts post-launch.

Due to tight launch timelines, we validated critical workflows with pilot users, iterating rapidly on feedback like confusing navigation and unclear AI responses. The numbers below come from pilot telemetry and post-launch surveys.

🚀 Workflow streamlining

Cut redundant tasks

40%

Centralizing AI chat, self-serve tiles, and sales tools cut redundant tasks like order tracking by 40%, letting employees focus on high-value work.

✂️ Platform consolidation

Eliminated daily platform-switching

72%

Integrated sales portal resources eliminated daily platform-switching for 72% of pilot users, freeing up time for higher-impact work.

💡 AI adoption

Feature discovery up, friction down

55% / 85%

Feature discovery rose by 55%. 85% of users reported "easier access to critical tools" once the new workspace was live.

Learnings

What I'll take into the next AI project.

The constraints on this project (tight deadlines, leadership-driven requirements, ambiguous AI capabilities) weren't problems to solve. They were the conditions of the work. Three things I learned about designing inside them.

⚖️

Business needs and user needs aren't opposites

Strict leadership requirements often mirrored hidden user needs, like faster workflows.

Treat constraints as guardrails, not roadblocks. Find the user need inside the business ask.

⏳

Limited time means ruthless prioritization

With tight deadlines, I prototyped only critical flows first like order tracking, then iterated post-launch.

Agility and quality coexist if you're honest about what to cut.

🤝

Compromise is strategy, not failure

Letting go of "perfect" features like custom animations freed time to solve bigger issues like AI response accuracy.

Progress beats perfection. Choose where to invest the polish.

Retrospective

What I took from this.

Two layers of impact that don't show up in the metrics. The work that shaped how I designed it, and the patterns the team kept using after I shipped.

Impact on me

Designing for AI trust is its own discipline.

Adoption didn't move when the AI got smarter. It moved when reps could see where each answer came from. Source-tagging, prompt suggestions, and recovery paths did more for behavior than any single model improvement. That changed how I scope every AI feature now.

14 enterprise-sales interviews shaped the workspace IA
5 prototype rounds with the pilot cohort before launch
What I'd do differently: invest in observability earlier, the late surveys made iteration slower than it needed to be

Impact on process

The patterns the next AI team picked up.

IntelliAssist 2.0 was the first full-screen AI surface at Dell. The patterns we shipped became the reference for every team that followed, so the next AI projects didn't have to rediscover what worked.

Source-tagged answer pattern adopted across Dell's enterprise AI surfaces
Prompt-suggestion grid added to the internal sales design system
Pilot feedback loop (telemetry plus surveys) reused on the next two AI launches