Equinix Intelligence
Gen AI-Powered Enterprise Support Chatbot
Designed an AI conversational assistant that reduced support costs by $1.8M annually while achieving 4.3/5 user satisfaction and 83% trust rate among technical users.
Case Study | Equinix
Product Overview
Equinix provides critical infrastructure services - data centers, network connectivity, and cloud interconnection - to enterprise customers worldwide. Their customer portal serves thousands of network engineers, IT managers, and cloud architects who manage complex infrastructure deployments.
The Problem
Challenge: Enterprise customers couldn't get quick answers to routine questions, leading to support bottlenecks, customer frustration, and escalating costs.
Role
Product Designer (Lead)
timeframe
6 Months
Team
Cross-functional
UX, Product, Engineering, CS
category
Research, Strategy, Design, Validation
Key Pain Points
​
-
Long wait times: 10-15 minutes average for simple support questions like order status checks
-
Fragmented documentation: Information scattered across docs.equinix.com, deploy.equinix.com, and internal systems
-
No self-service for account data: Customers had to contact support for basic order status or billing questions
-
High support costs: 8,500+ monthly tickets for routine, automatable queries
Business Impact
8500+
Monthly Support Tickets
10 - 15 Min
Avg Response Time
68%
Automable Queries
"I shouldn't need to contact support just to find out if my order is ready. This should be instant." - Network Engineer, User Interview
Research & Discovery
I led a comprehensive research initiative to understand user needs, pain points, and attitudes toward AI-powered support before designing any solutions.
15 Customer Interviews
Network engineers, IT managers, cloud architects across industries
8 Support Agent Interviews
GSD frontline agents, technical specialists, CSMs
10000+ Tickets Analyzed
GSD frontline agents, technical specialists, CSMs
Competitive Analysis
Benchmarked against Salesforce, Zendesk, Intercom, AWS
Key Research Insights
Critical Finding: Technical users will only trust AI if it shows sources and makes escalation to human experts easy. Paradoxically, making escalation prominent INCREASES AI adoption because users feel safe to try.
1. The Trust Paradox
Users are skeptical of AI accuracy yet frustrated by slow human support. They want AI speed but human reliability. Solution: Mandatory source attribution and one-click escalation.
​
2. Context is King
Users expect AI to understand their account, role, and current task without explaining everything. Generic responses feel irrelevant.
​
3. Transactional Data is the Differentiator
Users can Google general information. What they need is "Is MY order ready?" and "Why is MY bill higher?" Account-specific queries drive 65% of support volume.
​
4. Technical Users Demand Precision
Enterprise infrastructure users have zero tolerance for dumbed-down explanations or incorrect terminology. Technical accuracy > conversational personality.
"If an AI gives me one wrong answer, I'll never trust it again. But if it shows me where the information came from, I can verify and feel confident." - Cloud Architect, User Interview
Strategy & Definition
User Personas
Synthesized research into 3 primary personas to guide design decisions:
Sam - Technical Operations Engineer
High-frequency user. Needs instant answers for order status, configurations, documentation. Values speed and accuracy. Represents 55% of portal users.
Jordan - IT Manager
Moderate user. Focuses on billing, account management, strategic planning. Needs business-context answers. Represents 30% of users.
Alex - Cloud Architect
Occasional but high-value user. Needs deep technical specs, architecture guidance. Values comprehensive information. Represents 15% of users.
Feature Prioritization
Evaluated 13+ potential features using impact × effort framework. Focused MVP on 8 high-impact features.
MVP Core Features (Phase 1)
-
Natural Language Query Processing - Understand technical questions in plain English
-
Order Status Queries - Real-time lookup with detailed breakdowns
-
Billing Explanations - Charge breakdowns by service/usage with calculations
-
Documentation Search - Semantic search across all Equinix docs
-
Source Attribution - Every response shows where information came from
-
One-Click Escalation - Instant connection to appropriate human expert
-
Contextual Intelligence - Page-aware, account-aware assistance
-
Multi-Turn Conversations - Remember context across exchanges
Roadmap (Phase 2 & 3)
Deferred to post-launch: Multimodal support, proactive notifications, automated actions, mobile optimization, multi-language support.
Design Solution
Design Principles
1. Trust Through Transparency
Always show sources. Users should never question where information came from.
2. Technical Precision
Respect users' expertise. Use correct terminology, show exact data.
3. Progressive Disclosure
Start concise, offer depth on demand. Don't overwhelm.
4. Contextual Intelligence
Understand where user is and what they're doing. Reduce cognitive load.
5. Graceful Degradation
When AI can't help, make path to human support obvious and frictionless.
Key Features Designed
1. Conversational Chat Interface
-
Persistent widget accessible from any page in portal
-
Natural language processing for technical queries
-
Multi-turn conversations with context memory
-
Contextual suggestions based on current page
Chat Widget - Welcome Screen
Multi-turn Conversation
Mobile Chat Interface
Responsive Design (360 px width)
2. Transactional Data Integration
-
Order Status: Instant lookup with detailed item breakdown, completion estimates
-
Billing Queries: Charge explanations by service/cage with usage calculations
-
Service Health: Real-time monitoring status for user's specific services
-
Configuration Details: Account-specific settings and parameters
Order Status Query
Billing Explanation
Service Health Check
Complex Query - Multiple Orders Comparison
3. Trust Mechanisms
-
Source Attribution: Expandable panel showing documentation links, system data sources
-
Confidence Indicators: AI signals certainty level ("I'm confident" vs. "Based on available info...")
-
Limitations Acknowledgment: Explicit about what it can't do, offers human escalation
-
Easy Escalation: "Speak to an agent" button in every response, shows wait time
Documentation Search
Graceful Limitation Handling
Live Agent Transfer
4. Contextual Intelligence
-
Page Awareness: Knows what page user is on, assumes questions relate to current context
-
Account Awareness: Accesses user's specific orders, billing, services
-
Role-Based Responses: Tailors technical depth based on user role and permissions
-
Entry Point Suggestions: "Ask about this order" prompts on relevant pages
Contextual AI Help - Form Field Assistance
Chat Widget - API Authentication Guide
Testing & Validation - Usability Testing
Conducted moderated usability testing with 12 participants (mix of personas) using high-fidelity interactive prototype.
Test Results
87%
Success Rate
4.3/5
Satisfaction Score
82.5
SUS Score
92%
Would use in production
Key Findings from Testing
-
Order status queries: 100% success rate - Users loved instant, detailed responses
-
Source attribution builds trust: 83% clicked to verify sources (shows engagement)
-
Easy escalation reduces anxiety: Cited as #1 reason users felt comfortable trying AI
-
Technical precision appreciated: 89% rated AI as "appropriately technical for audience"
-
Contextual help needs prominence: 25% missed contextual help icons (design refined)
"This is exactly what I need. Fast, accurate, and I can verify the source. This would save me hours per week." - Operations Engineer, Usability Testing
What Worked Well
-
Deep user research - 15 interviews + ticket analysis gave us conviction in design decisions
-
Stakeholder alignment - Weekly reviews kept leadership and cross-functional team on same page
-
Iterative testing - Usability testing caught critical issues (contextual help visibility) before launch
-
Trust-first approach - Source attribution + escalation drove adoption beyond targets
-
Technical implementation partnership - Close collaboration with engineering on RAG architecture
.png){kind=icon}