AI-Powered Agent Companion
Transforming Telecom Support from Cost Center to Revenue Driver
Note: Because of the non-disclosure agreement, some of the information on this page has been replaced with fictitious ones or has been blurred out
Core product team responsible for decisions
Problem: Call quality crisis
Problem: Confusing UI is actively stealing 45% time from the human interaction.
The call center is the Final Safety Net for retention.
Secondary research: Observing agents' behaviour and analyzing data
Top findings from Secondary research
Workshop: Solution mapping
Participants: UX (Lead), AI Engineering, Data Architect, and Call Center Floor Manager.
The Problem: Finding the AI capabilities and limitations.
The Strategy: Not reworking the existing UI
The Outcome: We determined that manual data retrieval was the primary bottleneck. AI was selected to automate the ‘Information Hunt,’ allowing agents to focus 100% on the customer’s emotional state.
Why Not Just Redesign the UI?
A full UI redesign would take 2 years and wouldn't solve for the 'Information Scavenging' problem.
I proposed an AI-first layer to solve for AHT and Retention immediately while the legacy systems remain intact.
To-be journey for service design
A/B Testing
Creating and testing the concepts live
Churn risk indicator concepts
I found that seeing a 'Red' risk score wasn't enough (Challenge 1: Trust Gap).
Agents needed to know why the risk was high to believe the AI.
Learnings from testing
1. The Lag Challenge (Technical): AI took 3 seconds to load.
2. The Trust Gap (Psychological): Most agents ignored AI warnings.
Feedback from agents
3. Most agents requested help to navigate difficult conversations
4. Many agent miss cross-sell opportunities while manually calculating complex orders.
5. Many agents reported that they retype conversations after the call. This increases After-Call Work (ACW).
Final designs
AI assists the agents in all the functions
Launched Designs
Overcoming Technical & Psychological Hurdles
Challenge 1: The Trust Gap (Psychological)
Problem: Agents ignored AI churn warnings due to low trust.
Solution: Introduced “why” context (e.g., billing disputes), raising trust by 40%.
Building trust through transparency by adding reasons for AI predictions
Challenge 2 : The Lag Challenge (Technical):
Problem: MVP showed a 3-second latency for AI processing.
Solution: To prevent "dead air" on the call, I designed a Progressive Loading Sequence.
Progressive loading for overcoming AI latency in the final launch
Challenge 3: Helping agents navigate difficult conversations
Problem: Agents needed polished script to speak to customers
Solution: Low cognitive load, AI generated talking points that agents can use
Challenge 4: Identifying and capturing sales opportunities
Problem: Agents miss cross-sell opportunities while manually calculating complex orders.
Solution: An AI engine that instantly surfaces high-propensity offers with reasoning.
Challenge 5: Automating post-call documentation
Problem: Agents retype conversations after the call. This increases After-Call Work (ACW).
Solution: Structured AI generated CRM ready transcript.
Results
Quantifiable Impact on Business & People