Problem
Help sales/customer care representatives receive useful AI guidance during live conversations without increasing cognitive load.
Outcome
We built AI assistance system with a playbook-driven conversation graph that used CRM context, emotion, and user profile data to dynamically adapt guidance while also reducing operational cost.
What is Salesken’s Real-Time Assistant?
A system designed to help sales representatives navigate conversations, surface relevant information, and receive contextual recommendations without disrupting the flow of a live call during the live call.
My role
Lead Product Designer
2023–2025
Teams
Design + Engineering + QA
13
Features
~30%
Cost Reduction
~$65K+
Savings/month
24
Resolutions
4.8K+
Conversations/month
Impact after deployed
So the system needed to answer some questions
When should recommendations appear? | What information deserves attention? | How much information is too much? | When should AI stay silent? | How do we build trust in AI recommendations?
This turns the problem from interface design into attention and decision-support design.
The user’s Core problem.
Traditional workflows forced users to switch between multiple tools, documents, and dashboards while managing conversations in real time.
1. Listen actively
2. Build rapport
3. Follow process requirements
4. Handle objections
5. Capture information
6. Search for relevant resources
Product Design
Workflow Architecture
AI Interaction Design
Design System
Engineering Collaboration
Feature Design
​Edge Case Solutions
What was my contribution?
Understanding the Workflow:
A system designed to help sales representatives navigate conversations, surface relevant information, and receive contextual recommendations without disrupting the flow of a live call during the live call.
The system interacted with : QA/PCA | CRM Integrations | Admin Systems | Desktop Applications | Mobile Applications
Design decisions needed to support the broader ecosystem while remaining useful during live conversations.
Reduce Cognitive Load
Information must be scannable in seconds.
Build Trust Gradually
AI recommendations should feel assistive rather than intrusive.
Prioritize Actions Over Data
Representatives need guidance, not analytics.
Surface Information Only When Necessary
Users should never feel overwhelmed by recommendations.
Design Principles
Key Design Decisions:
Decision 01
Progressive Information Disclosure
Rather than showing everything immediately, information was revealed based on context and urgency.
Decision 02
Recommendation Prioritization
High-value recommendations received stronger visual emphasis while lower-priority insights remained available without competing for attention.
Decision 03
Workflow Consistency
Interaction patterns were standardized across multiple RTC experiences to reduce learning effort.
Outcomes
AI-assisted sales conversations | Reduced workflow fragmentation | Faster access to relevant information | More consistent user experiences | Improved integration with the wider Salesken ecosystem
Key Takeaway
Designing RTC reinforced a simple lesson: The most difficult challenge in AI products is not generating intelligence. It is deciding when, where, and how that intelligence should appear within a user's workflow.
Would you like to know more?
These projects offer just a glimpse into the creative process behind them, not the case studies. To learn more, feel free to drop me an email. I'm usually quick to respond, typically within a day.