AI Assistant Design for Pulse CRM
Natalia Calderón
Like this project
Posted May 26, 2025
Designed Pulse, an AI CRM assistant that resolves 50% of routine queries in minutes—cutting wait times from 48 hours to 5 while maintaining 80%+ satisfaction.
Likes
1
Views
0
Timeline
Feb 26, 2025 - Mar 26, 2025
Clients
Pulse
For: CRM Platform | Role: Lead UX Designer (Freelance) | Duration: 1 Month
🚩 The Problem
Customer support teams were stuck answering the same basic questions every day:
“Where’s the pricing tool?”
“How do I find a contract?”
Knowledge existed—but no one could find it. Response times averaged 48 hours, wasting time and delaying real customer issues.
🎯 Goal
Design a built-in AI assistant to:
Resolve 50% of routine CRM queries
Cut wait time from 48 hours to under 5 minutes
Maintain 80%+ user satisfaction
Free up CS teams for more strategic support
🔍 Research & Strategy
We studied 4 existing AI agents (HubSpot, Zendesk, Intercom, Freshsales) to define Pulse’s own approach:
🤖 Assistant, not Automator: Pulse suggests—never acts without the user
💬 Always Explain: Every suggestion includes a rationale
🧍♂️ User First: “Escalate to Support” is always one click away
Users didn’t want “AI.” They wanted instant, accurate help without the wait.
🧩 Defining Pulse’s Role
Pulse became a hybrid of:
A beginner’s guide
A responsive teammate
It answered common questions and assisted with tasks like:
“Show me overdue invoices”
“Take me to the bulk email editor”
“Use the enterprise upsell template”
Always helpful, never pushy—just like a great colleague.
🧪 Prototyping the Right Experience
🤖 Chat Design
❌ Complex Menus → Confusing, added friction
✅ Single Input Field → Natural, fast, intuitive
📍 Pulse’s Home in the UI
❌ Floating widget → Blocked CRM data
❌ Lateral nav → Cluttered main tools
✅ Top Menu Placement → Matched user mental models (like notifications/help)
🧷 Building Trust
To increase confidence, we added:
Explainability: “Recommended template based on: Client Tier (Enterprise) + Deal History”
User Control: Edit or reject suggestions anytime
Fail Gracefully: “I can’t find that—try rephrasing or contact support”
📊 Key Outcomes
We created a scalable, user-friendly assistant that:
Lives where users expect help
Feels like a smart teammate
Supports users without overstepping
“It’s like having the smartest CS rep in the room—always available, never annoying.”
📆 Next Steps
Wizard-of-Oz Testing with 10 enterprise users
Bias & Accuracy Audits for ethical AI scaling
Phased Rollout across 5 strategic client accounts
🧠 Key Learnings
🛠️ Designing AI Means Designing for Failure
Build-in fallback paths
Be transparent, not magical
Respect boundaries—AI should assist, not automate everything
🧱 The Interface Must Scale
Top-menu placement preps for future AI tools
Single-input design avoids UI bloat
Simpler ≠ Dumber—it just works better
💬 Takeaway
I don’t just design what looks good—I design what works, scales, and earns trust.0